Merge pull request #2221 from vladmandic/master

update dev

CHANGELOG.md

@@ -1,5 +1,13 @@
 # Change Log for SD.Next
 
+## Update for 2023-09-18
+
+Downgrade of `diffusers` to 0.20.2 due to critical issue with model offloading  
+This means that new model **Wuerstchen** is not supported until diffusers issue is resolved  
+
+- Added **change log** to UI, see *System -> Changelog*
+- **Extra networks**: faster search, ability to show/hide/sort networks
+
 ## Update for 2023-09-13
 
 Started as a mostly a service release with quite a few fixes, but then...  
@@ -22,11 +30,11 @@ Major changes how **hires** works as well as support for a very interesting new
 - diffusers:
   - allow loading of sd/sdxl models from safetensors without online connectivity
   - support for new model: [wuerstchen](https://huggingface.co/warp-ai/wuerstchen)  
-    its a high-resolution model (1024px+) that nearly doubls performance of sd-xl with much lower resource requirements  
+    its a high-resolution model (1024px+) thats ~40% faster than sd-xl with a bit lower resource requirements  
     go to *models -> huggingface -> search "warp-ai/wuerstchen" -> download*  
     its nearly 12gb in size, so be patient :)
 - minor re-layout of the main ui  
-- update **ui hints**  
+- updated **ui hints**  
 - updated **models -> civitai**  
   - search and download loras  
   - find previews for already downloaded models or loras  

extensions-builtin/LDSR/scripts/ldsr_model.py

@@ -2,8 +2,6 @@ import os
 import sys
 import traceback
 
-from basicsr.utils.download_util import load_file_from_url
-
 from modules.upscaler import Upscaler, UpscalerData
 from ldsr_model_arch import LDSR
 from modules import shared, script_callbacks
@@ -42,6 +40,7 @@ class UpscalerLDSR(Upscaler):
             print("Renaming model from model.pth to model.ckpt")
             os.rename(old_model_path, new_model_path)
 
+        from modules.modelloader import load_file_from_url
         if local_safetensors_path is not None and os.path.exists(local_safetensors_path):
             model = local_safetensors_path
         else:

extensions-builtin/Lora/lora.py

@@ -145,11 +145,11 @@ def assign_lora_names_to_compvis_modules(sd_model):
 
     sd_model.lora_layer_mapping = lora_layer_mapping
 
-def load_diffuser_lora(name, lora_on_disk, multiplier):
+def load_diffuser_lora(name, lora_on_disk, multiplier, num_loras):
     lora = LoraModule(name, lora_on_disk)
     lora.mtime = os.path.getmtime(lora_on_disk.filename)
     from modules.lora_diffusers import load_diffusers_lora
-    load_diffusers_lora(name, lora_on_disk, multiplier)
+    load_diffusers_lora(name, lora_on_disk, multiplier, num_loras)
     return lora
 
 
@@ -239,24 +239,36 @@ def load_loras(names, multipliers=None):
 
     failed_to_load_loras = []
 
+    recompile_model = False
+    if shared.opts.cuda_compile and shared.opts.cuda_compile_backend == "openvino_fx":
+        if len(names) == len(shared.compiled_model_state.lora_model):
+            for i, name in enumerate(names):
+                if shared.compiled_model_state.lora_model[i] != f"{name}:{multipliers[i]}":
+                    recompile_model = True
+                    break
+        else:
+            recompile_model = True
+        shared.compiled_model_state.lora_model = []
+    if recompile_model:
+        sd_models.unload_model_weights(op='model')
+        shared.opts.cuda_compile = False
+        sd_models.reload_model_weights(op='model')
+        shared.opts.cuda_compile = True
+
     for i, name in enumerate(names):
         lora = already_loaded.get(name, None) if shared.backend == shared.Backend.ORIGINAL else None
-
         lora_on_disk = loras_on_disk[i]
-
         if lora_on_disk is not None:
             if lora is None or os.path.getmtime(lora_on_disk.filename) > lora.mtime:
                 try:
                     if shared.backend == shared.Backend.DIFFUSERS:
-                        lora = load_diffuser_lora(name, lora_on_disk, multipliers[i] if multipliers else 1.0)
+                        lora = load_diffuser_lora(name, lora_on_disk, multipliers[i] if multipliers else 1.0, len(names))
                     else:
                         lora = load_lora(name, lora_on_disk)
                 except Exception as e:
                     errors.display(e, f"loading Lora {lora_on_disk.filename}")
                     continue
-
             lora.mentioned_name = name
-
             lora_on_disk.read_hash()
 
         if lora is None:
@@ -270,6 +282,10 @@ def load_loras(names, multipliers=None):
     if len(failed_to_load_loras) > 0:
         sd_hijack.model_hijack.comments.append("Failed to find Loras: " + ", ".join(failed_to_load_loras))
 
+    if recompile_model:
+        shared.log.info("Lora: Recompiling model")
+        sd_models.compile_diffusers(shared.sd_model)
+
 
 def lora_calc_updown(lora, module, target):
     with torch.no_grad():

extensions-builtin/ScuNET/scripts/scunet_model.py

@@ -7,8 +7,6 @@ import numpy as np
 import torch
 from tqdm import tqdm
 
-from basicsr.utils.download_util import load_file_from_url
-
 import modules.upscaler
 from modules import devices, modelloader, script_callbacks
 from scunet_model_arch import SCUNet as net
@@ -121,6 +119,7 @@ class UpscalerScuNET(modules.upscaler.Upscaler):
     def load_model(self, path: str):
         device = devices.get_device_for('scunet')
         if "http" in path:
+            from modules.modelloader import load_file_from_url
             filename = load_file_from_url(url=self.model_url, model_dir=self.model_download_path, file_name="%s.pth" % self.name, progress=True)
         else:
             filename = path

extensions-builtin/SwinIR/scripts/swinir_model.py

@@ -2,8 +2,7 @@ import os
 import numpy as np
 import torch
 from PIL import Image
-from basicsr.utils.download_util import load_file_from_url
-from tqdm.rich import tqdm
+from rich.progress import Progress, TextColumn, BarColumn, TaskProgressColumn, TimeRemainingColumn, TimeElapsedColumn
 from swinir_model_arch import SwinIR as net
 from swinir_model_arch_v2 import Swin2SR as net2
 from modules import modelloader, devices, script_callbacks, shared
@@ -45,11 +44,13 @@ class UpscalerSwinIR(Upscaler):
 
     def load_model(self, path, scale=4):
         if "http" in path:
-            dl_name = "%s%s" % (self.model_name.replace(" ", "_"), ".pth")
+            from modules.modelloader import load_file_from_url
+            dl_name = "%s%s" % (self.model_name.replace(" ", "_"), ".pth") # pylint: disable=consider-using-f-string
             filename = load_file_from_url(url=path, model_dir=self.model_download_path, file_name=dl_name, progress=True)
         else:
             filename = path
         if filename is None or not os.path.exists(filename):
+            shared.log.error(f"Model failed loading: type=SwinIR model={filename}")
             return None
         model_v2 = net2(
             upscale=scale,
@@ -78,6 +79,8 @@ class UpscalerSwinIR(Upscaler):
             resi_connection="3conv",
         )
         pretrained_model = torch.load(filename)
+        shared.log.info(f"Model loaded: type=SwinIR model={filename}")
+
         for model in [model_v1, model_v2]:
             for param in ["params_ema", "params", None]:
                 try:
@@ -140,7 +143,8 @@ def inference(img, model, tile, tile_overlap, window_size, scale):
     E = torch.zeros(b, c, h * sf, w * sf, dtype=devices.dtype, device=device_swinir).type_as(img)
     W = torch.zeros_like(E, dtype=devices.dtype, device=device_swinir)
 
-    with tqdm(total=len(h_idx_list) * len(w_idx_list), desc="Upscaling SwinIR") as pbar:
+    with Progress(TextColumn('[cyan]{task.description}'), BarColumn(), TaskProgressColumn(), TimeRemainingColumn(), TimeElapsedColumn(), console=shared.console) as progress:
+        task = progress.add_task(description="Upscaling Initializing", total=len(h_idx_list) * len(w_idx_list))
         for h_idx in h_idx_list:
             if state.interrupted or state.skipped:
                 break
@@ -159,7 +163,7 @@ def inference(img, model, tile, tile_overlap, window_size, scale):
                 W[
                 ..., h_idx * sf: (h_idx + tile) * sf, w_idx * sf: (w_idx + tile) * sf
                 ].add_(out_patch_mask)
-                pbar.update(1)
+                progress.update(task, advance=1, description="Upscaling")
     output = E.div_(W)
 
     return output

html/locale_en.json

@@ -63,7 +63,7 @@
   {"id":"","label":"UI position","localized":"","hint":"Location of extra networks"},
   {"id":"","label":"cover","localized":"","hint":"cover full area"},
   {"id":"","label":"inline","localized":"","hint":"inline with all additional elelemtns (scrollable)"},
-  {"id":"","label":"sidebar","localized":"","hint":"sidebar on the right side of the screen"},  
+  {"id":"","label":"sidebar","localized":"","hint":"sidebar on the right side of the screen"},
   {"id":"","label":"UI height (%)","localized":"","hint":""},
   {"id":"","label":"UI sidebar width (%)","localized":"","hint":""},
   {"id":"","label":"UI card preview lazy loading","localized":"","hint":""},
@@ -431,7 +431,7 @@
   {"id":"","label":"Create text file next to every image with generation parameters","localized":"","hint":""},
   {"id":"","label":"Create JSON log file for each saved image","localized":"","hint":"Save image information to a JSON file"},
   {"id":"","label":"Save copy of image before doing face restoration","localized":"","hint":""},
-  {"id":"","label":"Save copy of image before applying highres fix","localized":"","hint":""},
+  {"id":"","label":"Save copy of image before applying hires","localized":"","hint":""},
   {"id":"","label":"Save copy of image before applying color correction","localized":"","hint":""},
   {"id":"","label":"Save copy of the inpainting greyscale mask","localized":"","hint":""},
   {"id":"","label":"Save copy of inpainting masked composite","localized":"","hint":""},
@@ -585,7 +585,7 @@
   {"id":"","label":"Enable attention slicing","localized":"","hint":"Performs attention computation in steps instead of all at once. Slower inference times, but greatly reduced memory usage"},
   {"id":"","label":"Diffusers model loading variant","localized":"","hint":""},
   {"id":"","label":"Diffusers VAE loading variant","localized":"","hint":""},
-  {"id":"","label":"Diffusers LoRA loading variant","localized":"","hint":"'sequential apply' loads and applies each LoRA in order of appearance, 'merge and apply' loads all LoRAs and merges them in-memory before applying to model, 'diffusers default' uses single LoRA loading method"},
+  {"id":"","label":"Diffusers LoRA loading variant","localized":"","hint":"'sequential apply' loads and applies each LoRA in order of appearance, 'merge and apply' loads all LoRAs and merges them in-memory before applying to model, 'diffusers' uses diffusers default LoRA loading method"},
   {"id":"","label":"Torch inference mode","localized":"","hint":"Use torch inference mode"},
   {"id":"","label":"inference-mode","localized":"","hint":"Use torch.inference_mode"},
   {"id":"","label":"no-grad","localized":"","hint":"Use torch.no_grad"},

installer.py

@@ -429,8 +429,8 @@ def check_torch():
             torch_command = os.environ.get('TORCH_COMMAND', 'torch==2.0.0a0 intel_extension_for_pytorch==2.0.110+gitba7f6c1 -f https://developer.intel.com/ipex-whl-stable-xpu')
     elif allow_openvino and args.use_openvino:
         #Remove this after 2.1.0 releases
-        log.info('Using OpenVINO with Torch Nightly CPU')
-        torch_command = os.environ.get('TORCH_COMMAND', '--pre torch==2.1.0.dev20230713+cpu torchvision==0.16.0.dev20230713+cpu -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html')
+        log.info('Using OpenVINO')
+        torch_command = os.environ.get('TORCH_COMMAND', '--pre torch==2.1.0.dev20230726+cpu torchvision==0.16.0.dev20230726+cpu -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html')
     else:
         machine = platform.machine()
         if sys.platform == 'darwin':

javascript/extraNetworks.js

@@ -136,6 +136,31 @@ function saveCardDescription(event) {
   event.preventDefault();
 }
 
+async function filterExtraNetworksForTab(tabname, searchTerm) {
+  let found = 0;
+  let items = 0;
+  const t0 = performance.now();
+  const cards = Array.from(gradioApp().querySelectorAll(`#${tabname}_extra_tabs div.card`));
+  cards.forEach((elem) => {
+    items += 1;
+    if (searchTerm === '') {
+      elem.style.display = '';
+    } else {
+      let text = `${elem.querySelector('.name').textContent.toLowerCase()} ${elem.querySelector('.search_term').textContent}`;
+      text = text.toLowerCase().replace('models--', 'Diffusers').replace('\\', '/');
+      if (text.indexOf(searchTerm) === -1) {
+        elem.style.display = 'none';
+      } else {
+        elem.style.display = '';
+        found += 1;
+      }
+    }
+  });
+  const t1 = performance.now();
+  if (found > 0) log(`filterExtraNetworks: text=${searchTerm} items=${items} match=${found} time=${Math.round(1000 * (t1 - t0)) / 1000000}`);
+  else log(`filterExtraNetworks: text=all items=${items} time=${Math.round(1000 * (t1 - t0)) / 1000000}`);
+}
+
 function setupExtraNetworksForTab(tabname) {
   gradioApp().querySelector(`#${tabname}_extra_tabs`).classList.add('extra-networks');
   const tabs = gradioApp().querySelector(`#${tabname}_extra_tabs > div`);
@@ -157,14 +182,9 @@ function setupExtraNetworksForTab(tabname) {
   search.addEventListener('input', (evt) => {
     if (searchTimer) clearTimeout(searchTimer);
     searchTimer = setTimeout(() => {
-      const searchTerm = search.value.toLowerCase();
-      gradioApp().querySelectorAll(`#${tabname}_extra_tabs div.card`).forEach((elem) => {
-        let text = `${elem.querySelector('.name').textContent.toLowerCase()} ${elem.querySelector('.search_term').textContent.toLowerCase()}`;
-        text = text.replace('models--', 'Diffusers');
-        elem.style.display = text.indexOf(searchTerm) === -1 ? 'none' : '';
-      });
+      filterExtraNetworksForTab(tabname, search.value.toLowerCase());
       searchTimer = null;
-    }, 100);
+    }, 150);
   });
 
   let hoverTimer = null;
@@ -186,7 +206,10 @@ function setupExtraNetworksForTab(tabname) {
 
   const intersectionObserver = new IntersectionObserver((entries) => {
     if (!en) return;
-    for (const el of Array.from(gradioApp().querySelectorAll('.extra-networks-page'))) el.style.height = `${window.opts.extra_networks_height}vh`;
+    for (const el of Array.from(gradioApp().querySelectorAll('.extra-networks-page'))) {
+      el.style.height = `${window.opts.extra_networks_height}vh`;
+      el.parentElement.style.width = '-webkit-fill-available';
+    }
     if (entries[0].intersectionRatio > 0) {
       if (window.opts.extra_networks_card_cover === 'cover') {
         en.style.transition = '';
@@ -269,6 +292,7 @@ function tryToRemoveExtraNetworkFromPrompt(textarea, text) {
 }
 
 function refreshExtraNetworks(tabname) {
+  console.log('refreshExtraNetworks', tabname, gradioApp().querySelector(`#${tabname}_extra_networks textarea`)?.value);
   gradioApp().querySelector(`#${tabname}_extra_networks textarea`)?.dispatchEvent(new Event('input'));
 }
 

javascript/midnight-barbie.css

@@ -314,4 +314,4 @@ svg.feather.feather-image, .feather .feather-image { display: none }
   --size-9: 64px;
   --size-14: 64px;
 }
-/*Midnight-Barbie, By Nyxxia*/
+/*Midnight-Barbie, By Nyxxia*/

javascript/style.css

@@ -249,10 +249,10 @@ table.settings-value-table td { padding: 0.4em; border: 1px solid #ccc; max-widt
 .extra-network-cards .card:hover .overlay .tags { display: block; }
 .extra-network-cards .card:hover .overlay .description { display: block; }
 .extra-network-cards .card:hover .preview { box-shadow: none; filter: grayscale(100%); }
-#txt2img_description, #img2img_description { max-height: 63px; overflow-y: auto !important; } 
-#txt2img_description > label > textarea, #img2img_description > label > textarea { font-size: 0.9em } 
+#txt2img_description, #img2img_description { max-height: 63px; overflow-y: auto !important; }
+#txt2img_description > label > textarea, #img2img_description > label > textarea { font-size: 0.9em }
 
-/* controlnet */
+/* controlnet
 .controlnet_control_type .controlnet_control_type_filter_group .wrap:last-of-type { display: grid; grid-auto-flow: row; grid-template-columns: repeat(4, minmax(0, 1fr)); }
 fieldset.controlnet_resize_mode_radio .wrap:last-of-type, fieldset.controlnet_control_mode_radio .wrap:last-of-type { flex-direction: column; }
 div.controlnet_preprocessor_model { display: grid; grid-auto-flow: row; grid-template-columns: 1fr max-content; }
@@ -263,6 +263,7 @@ div.controlnet_image_controls { display: grid; grid-template-columns: repeat(4,
 div.controlnet_image_controls .controlnet_invert_warning { grid-column: 1 / -1; }
 div.controlnet_image_controls button { justify-self: center; }
 div.controlnet_main_options { display: grid; grid-template-columns: 1fr 1fr; grid-auto-flow: row; }
+ */
 
 /* specific elements */
 #modelmerger_interp_description { margin-top: 1em; margin-bottom: 1em; }
@@ -281,6 +282,12 @@ div.controlnet_main_options { display: grid; grid-template-columns: 1fr 1fr; gri
 .log-monitor { display: none; justify-content: unset !important; overflow: hidden; padding: 0; margin-top: auto; font-family: monospace; font-size: 0.85em; }
 .log-monitor td, .log-monitor th { padding-left: 1em; }
 
+/* changelog */
+.md h2 { background-color: var(--background-fill-primary); padding: 0.5em; }
+.md ul { list-style-type: square !important; text-indent: 1em; margin-left: 4em; }
+.md li { list-style-position: outside !important; text-indent: 0; }
+.md p { margin-left: 2em; }
+
 /* custom component */
 .folder-selector textarea { height: 2em !important; padding: 6px !important; }
 
@@ -295,6 +302,7 @@ div.controlnet_main_options { display: grid; grid-template-columns: 1fr 1fr; gri
 .loader::before { border-top-color: var(--primary-900); animation: 3s spin linear infinite; }
 .loader::after { border-top-color: var(--primary-300); animation: spin 1.5s linear infinite; }
 
+
 @keyframes move {
   from { background-position-x: 0, -40px; }
   to { background-position-x: 0, 40px; }

modules/api/api.py

@@ -2,7 +2,7 @@ import io
 import time
 import base64
 from io import BytesIO
-from typing import List, Dict, Any
+from typing import List, Dict, Any, Optional
 from threading import Lock
 from secrets import compare_digest
 from fastapi import FastAPI, APIRouter, Depends
@@ -148,6 +148,7 @@ class Api:
         self.add_api_route("/sdapi/v1/scripts", self.get_scripts_list, methods=["GET"], response_model=models.ScriptsList)
         self.add_api_route("/sdapi/v1/script-info", self.get_script_info, methods=["GET"], response_model=List[models.ScriptInfo])
         self.add_api_route("/sdapi/v1/log", self.get_log_buffer, methods=["GET"], response_model=List) # bypass auth
+        self.add_api_route("/sdapi/v1/extra-networks", self.get_extra_networks, methods=["GET"], response_model=List[models.ExtraNetworkItem])
         self.default_script_arg_txt2img = []
         self.default_script_arg_img2img = []
 
@@ -180,10 +181,12 @@ class Api:
         i2ilist = [script.name for script in scripts.scripts_img2img.scripts if script.name is not None]
         return models.ScriptsList(txt2img = t2ilist, img2img = i2ilist)
 
-    def get_script_info(self):
+    def get_script_info(self, script_name: Optional[str] = None):
         res = []
         for script_list in [scripts.scripts_txt2img.scripts, scripts.scripts_img2img.scripts]:
-            res += [script.api_info for script in script_list if script.api_info is not None]
+            for script in script_list:
+                if script.api_info is not None and (script_name is None or script_name == script.api_info.name):
+                    res.append(script.api_info)
         return res
 
     def get_script(self, script_name, script_runner):
@@ -500,6 +503,33 @@ class Api:
             "skipped": convert_embeddings(db.skipped_embeddings),
         }
 
+    def get_extra_networks(self, page: Optional[str] = None, name: Optional[str] = None, filename: Optional[str] = None, title: Optional[str] = None, fullname: Optional[str] = None, hash: Optional[str] = None): # pylint: disable=redefined-builtin
+        res = []
+        for pg in shared.extra_networks:
+            if page is not None and pg.name != page.lower():
+                continue
+            for item in pg.items:
+                if name is not None and item.get('name', '') != name:
+                    continue
+                if title is not None and item.get('title', '') != title:
+                    continue
+                if filename is not None and item.get('filename', '') != filename:
+                    continue
+                if fullname is not None and item.get('fullname', '') != fullname:
+                    continue
+                if hash is not None and (item.get('shorthash', None) or item.get('hash')) != hash:
+                    continue
+                res.append({
+                    'name': item.get('name', ''),
+                    'type': pg.name,
+                    'title': item.get('title', None),
+                    'fullname': item.get('fullname', None),
+                    'filename': item.get('filename', None),
+                    'hash': item.get('shorthash', None) or item.get('hash'),
+                    "preview": item.get('preview', None),
+                })
+        return res
+
     def refresh_checkpoints(self):
         return shared.refresh_checkpoints()
 

modules/api/models.py

@@ -272,6 +272,20 @@ class StyleItem(BaseModel):
     filename: Optional[str] = Field(title="Filename")
     preview: Optional[str] = Field(title="Preview")
 
+class ExtraNetworkItem(BaseModel):
+    name: str = Field(title="Name")
+    type: str = Field(title="Type")
+    title: Optional[str] = Field(title="Title")
+    fullname: Optional[str] = Field(title="Fullname")
+    filename: Optional[str] = Field(title="Filename")
+    hash: Optional[str] = Field(title="Hash")
+    preview: Optional[str] = Field(title="Preview image URL")
+    # description: Optional[str] = Field(title="Description")
+    # info: Optional[str] = Field(title="Information")
+    # metadata: Optional[Any] = Field(title="Metadata")
+    # local: Optional[str] = Field(title="Local")
+
+
 class ArtistItem(BaseModel):
     name: str = Field(title="Name")
     score: float = Field(title="Score")
@@ -303,7 +317,7 @@ class ScriptArg(BaseModel):
     minimum: Optional[Any] = Field(default=None, title="Minimum", description="Minimum allowed value for the argumentin UI")
     maximum: Optional[Any] = Field(default=None, title="Minimum", description="Maximum allowed value for the argumentin UI")
     step: Optional[Any] = Field(default=None, title="Minimum", description="Step for changing value of the argumentin UI")
-    choices: Optional[List[str]] = Field(default=None, title="Choices", description="Possible values for the argument")
+    choices: Optional[Any] = Field(default=None, title="Choices", description="Possible values for the argument")
 
 
 class ScriptInfo(BaseModel):

modules/devices.py

@@ -115,7 +115,7 @@ def torch_gc(force=False):
     if oom > previous_oom:
         previous_oom = oom
         shared.log.warning(f'GPU out-of-memory error: {mem}')
-    if used > 95:
+    if used > 90:
         shared.log.info(f'GPU high memory utilization: {used}% {mem}')
         force = True
     if not force:

modules/dml/__init__.py

@@ -9,7 +9,7 @@ default_memory_provider = "None"
 if platform.system() == "Windows":
     memory_providers.append("Performance Counter")
     default_memory_provider = "Performance Counter"
-do_nothing = lambda: None
+do_nothing = lambda: None # pylint: disable=unnecessary-lambda-assignment
 
 def _set_memory_provider():
     from modules.shared import opts, cmd_opts, log
@@ -63,7 +63,7 @@ def directml_init():
     return True, None
 
 def directml_do_hijack():
-    import modules.dml.hijack
+    import modules.dml.hijack # pylint: disable=unused-import
     from modules.devices import device
 
     if not torch.dml.has_float64_support(device):
@@ -79,9 +79,9 @@ class OverrideItem(NamedTuple):
     message: Optional[str]
 
 opts_override_table = {
-    "diffusers_generator_device": OverrideItem("cpu", None, "DirectML does not support torch Generator API."),
-    "diffusers_model_cpu_offload": OverrideItem(False, None, "Diffusers' model CPU offloading does not support DirectML devices."),
-    "diffusers_seq_cpu_offload": OverrideItem(False, lambda opts: opts.diffusers_pipeline != "Stable Diffusion XL", "Diffusers' sequential CPU offloading is available only on StableDiffusionXLPipeline with DirectML devices."),
+    "diffusers_generator_device": OverrideItem("cpu", None, "DirectML does not support torch Generator API"),
+    "diffusers_model_cpu_offload": OverrideItem(False, None, "Diffusers model CPU offloading does not support DirectML devices"),
+    "diffusers_seq_cpu_offload": OverrideItem(False, lambda opts: opts.diffusers_pipeline != "Stable Diffusion XL", "Diffusers sequential CPU offloading is available only on StableDiffusionXLPipeline with DirectML devices"),
 }
 
 def directml_override_opts():
@@ -96,11 +96,9 @@ def directml_override_opts():
         if getattr(shared.opts, key) != item.value and (item.condition is None or item.condition(shared.opts)):
             count += 1
             setattr(shared.opts, key, item.value)
-            if item.message is not None:
-                shared.log.warning(item.message)
-            shared.log.warning(f'{key} is automatically overriden to {item.value}.')
+            shared.log.warning(f'Overriding: {key}={item.value} {item.message if item.message is not None else ""}')
 
     if count > 0:
-        shared.log.info(f'{count} options are automatically overriden. If you want to keep them from overriding, run with --experimental argument.')
+        shared.log.info(f'Options override: count={count}. If you want to keep them from overriding, run with --experimental argument.')
 
     _set_memory_provider()

modules/dml/hijack/realesrgan_model.py

@@ -1,6 +1,6 @@
 import math
 import torch
-from realesrgan import RealESRGANer
+from modules.realesrgan_model_arch import RealESRGANer
 
 
 # DML Solution: Some of contents of output tensor turn to 0 after Extended Slices. Move it to cpu.

modules/esrgan_model.py

@@ -3,12 +3,12 @@ import os
 import numpy as np
 import torch
 from PIL import Image
-from basicsr.utils.download_util import load_file_from_url
+from rich.progress import Progress, TextColumn, BarColumn, TaskProgressColumn, TimeRemainingColumn, TimeElapsedColumn
 
 import modules.esrgan_model_arch as arch
 from modules import modelloader, images, devices
 from modules.upscaler import Upscaler, UpscalerData
-from modules.shared import opts
+from modules.shared import opts, log, console
 
 
 
@@ -152,6 +152,7 @@ class UpscalerESRGAN(Upscaler):
 
     def load_model(self, path: str):
         if "http" in path:
+            from modules.modelloader import load_file_from_url
             filename = load_file_from_url(
                 url=self.model_url,
                 model_dir=self.model_download_path,
@@ -161,10 +162,11 @@ class UpscalerESRGAN(Upscaler):
         else:
             filename = path
         if not os.path.exists(filename) or filename is None:
-            print(f"Unable to load {self.model_path} from {filename}")
+            log.error(f"Model failed loading: type=ESRGAN model={filename}")
             return None
 
         state_dict = torch.load(filename, map_location='cpu' if devices.device_esrgan.type == 'mps' else None)
+        log.info(f"Model loaded: type=ESRGAN model={filename}")
 
         if "params_ema" in state_dict:
             state_dict = state_dict["params_ema"]
@@ -216,16 +218,20 @@ def esrgan_upscale(model, img):
     newtiles = []
     scale_factor = 1
 
-    for y, h, row in grid.tiles:
-        newrow = []
-        for tiledata in row:
-            x, w, tile = tiledata
-
-            output = upscale_without_tiling(model, tile)
-            scale_factor = output.width // tile.width
-
-            newrow.append([x * scale_factor, w * scale_factor, output])
-        newtiles.append([y * scale_factor, h * scale_factor, newrow])
+    with Progress(TextColumn('[cyan]{task.description}'), BarColumn(), TaskProgressColumn(), TimeRemainingColumn(), TimeElapsedColumn(), console=console) as progress:
+        total = 0
+        for y, h, row in grid.tiles:
+            total += len(row)
+        task = progress.add_task(description="Upscaling", total=total)
+        for y, h, row in grid.tiles:
+            newrow = []
+            for tiledata in row:
+                x, w, tile = tiledata
+                output = upscale_without_tiling(model, tile)
+                scale_factor = output.width // tile.width
+                newrow.append([x * scale_factor, w * scale_factor, output])
+                progress.update(task, advance=1, description="Upscaling")
+            newtiles.append([y * scale_factor, h * scale_factor, newrow])
 
     newgrid = images.Grid(newtiles, grid.tile_w * scale_factor, grid.tile_h * scale_factor, grid.image_w * scale_factor, grid.image_h * scale_factor, grid.overlap * scale_factor)
     output = images.combine_grid(newgrid)

modules/gfpgan_model.py

@@ -29,12 +29,13 @@ def gfpgann():
         latest_file = max(models, key=os.path.getctime)
         model_file = latest_file
     else:
-        print("Unable to load gfpgan model!")
+        shared.log.error(f"Model failed loading: type=GFPGAN model={model_file}")
         return None
     if hasattr(facexlib.detection.retinaface, 'device'):
         facexlib.detection.retinaface.device = devices.device_gfpgan
     model = gfpgan_constructor(model_path=model_file, upscale=1, arch='clean', channel_multiplier=2, bg_upsampler=None, device=devices.device_gfpgan)
     loaded_gfpgan_model = model
+    shared.log.info(f"Model loaded: type=GFPGAN model={model_file}")
 
     return model
 

modules/hypernetworks/hypernetwork.py

@@ -221,10 +221,10 @@ class Hypernetwork:
             torch.save(optimizer_saved_dict, f"{filename}.optim")
 
     def load(self, filename):
-        self.filename = filename
+        self.filename = filename if os.path.exists(filename) else os.path.join(shared.opts.hypernetwork_dir, filename)
         if self.name is None:
-            self.name = os.path.splitext(os.path.basename(filename))[0]
-        with progress.open(filename, 'rb', description=f'Loading hypernetwork: [cyan]{filename}', auto_refresh=True, console=shared.console) as f:
+            self.name = os.path.splitext(os.path.basename(self.filename))[0]
+        with progress.open(self.filename, 'rb', description=f'Loading hypernetwork: [cyan]{self.filename}', auto_refresh=True, console=shared.console) as f:
             state_dict = torch.load(f, map_location='cpu')
         self.layer_structure = state_dict.get('layer_structure', [1, 2, 1])
         self.optional_info = state_dict.get('optional_info', None)

modules/images.py

@@ -273,7 +273,7 @@ re_nonletters = re.compile(r'[\s' + string.punctuation + ']+')
 re_pattern = re.compile(r"(.*?)(?:\[([^\[\]]+)\]|$)")
 re_pattern_arg = re.compile(r"(.*)<([^>]*)>$")
 max_filename_part_length = 128
-NOTHING_AND_SKIP_PREVIOUS_TEXT = object()
+NOTHING = object()
 
 
 def sanitize_filename_part(text, replace_spaces=True):
@@ -291,13 +291,13 @@ def sanitize_filename_part(text, replace_spaces=True):
 
 class FilenameGenerator:
     replacements = {
-        'batch_number': lambda self: NOTHING_AND_SKIP_PREVIOUS_TEXT if self.p is None or self.p.batch_size == 1 else self.p.batch_index + 1,
+        'batch_number': lambda self: NOTHING if self.index <= 1 else self.index,
         'cfg': lambda self: self.p and self.p.cfg_scale,
         'clip_skip': lambda self: self.p and self.p.clip_skip,
         'date': lambda self: datetime.datetime.now().strftime('%Y-%m-%d'),
         'datetime': lambda self, *args: self.datetime(*args),  # accepts formats: [datetime], [datetime<Format>], [datetime<Format><Time Zone>]
-        'denoising': lambda self: self.p.denoising_strength if self.p and self.p.denoising_strength else NOTHING_AND_SKIP_PREVIOUS_TEXT,
-        'generation_number': lambda self: NOTHING_AND_SKIP_PREVIOUS_TEXT if self.p.n_iter == 1 and self.p.batch_size == 1 else self.p.iteration * self.p.batch_size + self.p.batch_index + 1,
+        'denoising': lambda self: self.p.denoising_strength if self.p and self.p.denoising_strength else NOTHING,
+        'generation_number': lambda self: NOTHING if self.p.n_iter == 1 and self.p.batch_size == 1 else self.p.iteration * self.p.batch_size + self.p.batch_index + 1,
         'hasprompt': lambda self, *args: self.hasprompt(*args),  # accepts formats:[hasprompt<prompt1|default><prompt2>..]
         'height': lambda self: self.image.height,
         'image_hash': lambda self: self.image_hash(),
@@ -320,11 +320,12 @@ class FilenameGenerator:
     }
     default_time_format = '%Y%m%d%H%M%S'
 
-    def __init__(self, p, seed, prompt, image):
+    def __init__(self, p, seed, prompt, image, index = 0):
         self.p = p
         self.seed = seed
         self.prompt = prompt
         self.image = image
+        self.index = index if self.p is None or self.p.batch_size == 1 else self.p.batch_index + 1
 
     def hasprompt(self, *args):
         lower = self.prompt.lower()
@@ -403,7 +404,7 @@ class FilenameGenerator:
                 except Exception as e:
                     replacement = None
                     errors.display(e, 'filename pattern')
-                if replacement == NOTHING_AND_SKIP_PREVIOUS_TEXT:
+                if replacement == NOTHING:
                     continue
                 elif replacement is not None:
                     res += text + str(replacement)
@@ -498,7 +499,7 @@ save_thread = threading.Thread(target=atomically_save_image, daemon=True)
 save_thread.start()
 
 
-def save_image(image, path, basename, seed=None, prompt=None, extension='jpg', info=None, short_filename=False, no_prompt=False, grid=False, pnginfo_section_name='parameters', p=None, existing_info=None, forced_filename=None, suffix="", save_to_dirs=None):
+def save_image(image, path, basename, seed=None, prompt=None, extension='jpg', info=None, short_filename=False, no_prompt=False, grid=False, pnginfo_section_name='parameters', p=None, existing_info=None, forced_filename=None, suffix="", save_to_dirs=None, index=0):
     """Save an image.
     Args:
         image (`PIL.Image`):
@@ -536,7 +537,7 @@ def save_image(image, path, basename, seed=None, prompt=None, extension='jpg', i
         return None, None
     if path is None or len(path) == 0: # set default path to avoid errors when functions are triggered manually or via api and param is not set
         path = shared.opts.outdir_save
-    namegen = FilenameGenerator(p, seed, prompt, image)
+    namegen = FilenameGenerator(p, seed, prompt, image, index)
     if save_to_dirs is None:
         save_to_dirs = (grid and shared.opts.grid_save_to_dirs) or (not grid and shared.opts.save_to_dirs and not no_prompt)
     if save_to_dirs:

modules/intel/ipex/attention.py

@@ -64,8 +64,14 @@ def torch_bmm(input, mat2, *, out=None):
 original_scaled_dot_product_attention = torch.nn.functional.scaled_dot_product_attention
 def scaled_dot_product_attention(query, key, value, attn_mask=None, dropout_p=0.0, is_causal=False):
     #ARC GPUs can't allocate more than 4GB to a single block, Slice it:
-    shape_one, batch_size_attention, query_tokens, shape_four = query.shape
-    block_multiply = 2.4 if query.dtype == torch.float32 else 1.2
+    if len(query.shape) == 3:
+        batch_size_attention, query_tokens, shape_four = query.shape
+        shape_one = 1
+        no_shape_one = True
+    else:
+        shape_one, batch_size_attention, query_tokens, shape_four = query.shape
+        no_shape_one = False
+    block_multiply = 3.6 if query.dtype == torch.float32 else 1.8
     block_size = (shape_one * batch_size_attention * query_tokens * shape_four) / 1024 * block_multiply #MB
     split_slice_size = batch_size_attention
     if block_size >= 4000:
@@ -101,21 +107,39 @@ def scaled_dot_product_attention(query, key, value, attn_mask=None, dropout_p=0.
                 for i2 in range(query_tokens // split_2_slice_size): # pylint: disable=invalid-name
                     start_idx_2 = i2 * split_2_slice_size
                     end_idx_2 = (i2 + 1) * split_2_slice_size
-                    hidden_states[:, start_idx:end_idx, start_idx_2:end_idx_2] = original_scaled_dot_product_attention(
-                        query[:, start_idx:end_idx, start_idx_2:end_idx_2],
-                        key[:, start_idx:end_idx, start_idx_2:end_idx_2],
-                        value[:, start_idx:end_idx, start_idx_2:end_idx_2],
-                        attn_mask=attn_mask[:, start_idx:end_idx, start_idx_2:end_idx_2] if attn_mask is not None else attn_mask,
+                    if no_shape_one:
+                        hidden_states[start_idx:end_idx, start_idx_2:end_idx_2] = original_scaled_dot_product_attention(
+                            query[start_idx:end_idx, start_idx_2:end_idx_2],
+                            key[start_idx:end_idx, start_idx_2:end_idx_2],
+                            value[start_idx:end_idx, start_idx_2:end_idx_2],
+                            attn_mask=attn_mask[start_idx:end_idx, start_idx_2:end_idx_2] if attn_mask is not None else attn_mask,
+                            dropout_p=dropout_p, is_causal=is_causal
+                        )
+                    else:
+                        hidden_states[:, start_idx:end_idx, start_idx_2:end_idx_2] = original_scaled_dot_product_attention(
+                            query[:, start_idx:end_idx, start_idx_2:end_idx_2],
+                            key[:, start_idx:end_idx, start_idx_2:end_idx_2],
+                            value[:, start_idx:end_idx, start_idx_2:end_idx_2],
+                            attn_mask=attn_mask[:, start_idx:end_idx, start_idx_2:end_idx_2] if attn_mask is not None else attn_mask,
+                            dropout_p=dropout_p, is_causal=is_causal
+                        )
+            else:
+                if no_shape_one:
+                    hidden_states[start_idx:end_idx] = original_scaled_dot_product_attention(
+                        query[start_idx:end_idx],
+                        key[start_idx:end_idx],
+                        value[start_idx:end_idx],
+                        attn_mask=attn_mask[start_idx:end_idx] if attn_mask is not None else attn_mask,
+                        dropout_p=dropout_p, is_causal=is_causal
+                    )
+                else:
+                    hidden_states[:, start_idx:end_idx] = original_scaled_dot_product_attention(
+                        query[:, start_idx:end_idx],
+                        key[:, start_idx:end_idx],
+                        value[:, start_idx:end_idx],
+                        attn_mask=attn_mask[:, start_idx:end_idx] if attn_mask is not None else attn_mask,
                         dropout_p=dropout_p, is_causal=is_causal
                     )
-            else:
-                hidden_states[:, start_idx:end_idx] = original_scaled_dot_product_attention(
-                    query[:, start_idx:end_idx],
-                    key[:, start_idx:end_idx],
-                    value[:, start_idx:end_idx],
-                    attn_mask=attn_mask[:, start_idx:end_idx] if attn_mask is not None else attn_mask,
-                    dropout_p=dropout_p, is_causal=is_causal
-                )
     else:
         return original_scaled_dot_product_attention(
             query, key, value, attn_mask=attn_mask, dropout_p=dropout_p, is_causal=is_causal

modules/intel/openvino/__init__.py

@@ -1,30 +1,53 @@
 import os
 import torch
-from openvino.frontend.pytorch.torchdynamo.execute import execute, partitioned_modules, compiled_cache
+from openvino.frontend import FrontEndManager
+from openvino.frontend.pytorch.fx_decoder import TorchFXPythonDecoder
 from openvino.frontend.pytorch.torchdynamo.partition import Partitioner
-from openvino.runtime import Core, Type, PartialShape
+from openvino.runtime import Core, Type, PartialShape, serialize
 from torch._dynamo.backends.common import fake_tensor_unsupported
 from torch._dynamo.backends.registry import register_backend
 from torch.fx.experimental.proxy_tensor import make_fx
-from torch._inductor.compile_fx import compile_fx
+from torch.utils._pytree import tree_flatten
+from types import MappingProxyType
 from hashlib import sha256
+import functools
 from modules import shared, devices
 
-@register_backend
-@fake_tensor_unsupported
-def openvino_fx(subgraph, example_inputs):
-    executor_parameters = None
+compiled_cache = {}
+max_openvino_partitions = 0
+partitioned_modules = {}
+
+DEFAULT_OPENVINO_PYTHON_CONFIG = MappingProxyType(
+    {
+        "use_python_fusion_cache": True,
+        "allow_single_op_fusion": True,
+    },
+)
+
+class OpenVINOGraphModule(torch.nn.Module):
+    def __init__(self, gm, partition_id, use_python_fusion_cache, model_hash_str: str = None, file_name=""):
+        super().__init__()
+        self.gm = gm
+        self.partition_id = partition_id
+        self.executor_parameters = {"use_python_fusion_cache": use_python_fusion_cache,
+                                    "model_hash_str": model_hash_str}
+        self.file_name = file_name
+        self.perm_fallback = False
+
+    def __call__(self, *args):
+        #if self.perm_fallback:
+        #    return self.gm(*args)
+
+        #try:
+        result = openvino_execute(self.gm, *args, executor_parameters=self.executor_parameters, partition_id=self.partition_id, file_name=self.file_name)
+        #except Exception:
+        #    self.perm_fallback = True
+        #    return self.gm(*args)
+
+        return result
+
+def get_device():
     core = Core()
-    if os.getenv("OPENVINO_TORCH_MODEL_CACHING") != "0":
-        os.environ.setdefault('OPENVINO_TORCH_MODEL_CACHING', "1")
-        model_hash_str = sha256(subgraph.code.encode('utf-8')).hexdigest()
-        executor_parameters = {"model_hash_str": model_hash_str}
-
-    example_inputs.reverse()
-    cache_root = "./cache/"
-    if os.getenv("OPENVINO_TORCH_CACHE_DIR") is not None:
-        cache_root = os.getenv("OPENVINO_TORCH_CACHE_DIR")
-
     if os.getenv("OPENVINO_TORCH_BACKEND_DEVICE") is not None:
         device = os.getenv("OPENVINO_TORCH_BACKEND_DEVICE")
     elif any(openvino_cpu in cpu_module.lower() for cpu_module in shared.cmd_opts.use_cpu for openvino_cpu in ["openvino", "all"]):
@@ -43,83 +66,316 @@ def openvino_fx(subgraph, example_inputs):
     os.environ.setdefault('OPENVINO_TORCH_BACKEND_DEVICE', device)
     shared.log.debug(f"OpenVINO Device: {device}")
     if shared.opts.cuda_compile_errors and device not in core.available_devices:
-        shared.log.warning(f"OpenVINO: Specified device {device} is not in the list of OpenVINO Available Devices")
+        shared.log.error(f"OpenVINO: Specified device {device} is not in the list of OpenVINO Available Devices")
     assert device in core.available_devices, f"OpenVINO: Specified device {device} is not in the list of OpenVINO Available Devices"
 
-    #Cache saving keeps increasing the partition id
-    #This loop check if non 0 partition id caches exist
-    #Takes 0.002 seconds when nothing is found
-    use_cached_file = False
-    for i in range(100):
-        file_name = get_cached_file_name(*example_inputs, model_hash_str=str(model_hash_str + str(i)), device=device, cache_root=cache_root)
-        if file_name is not None and os.path.isfile(file_name + ".xml") and os.path.isfile(file_name + ".bin"):
-            use_cached_file = True
-            break
+    return device
 
-    if use_cached_file:
-        om = core.read_model(file_name + ".xml")
+def cache_root_path():
+    cache_root = "./cache/"
+    if os.getenv("OPENVINO_TORCH_CACHE_DIR") is not None:
+        cache_root = os.getenv("OPENVINO_TORCH_CACHE_DIR")
+    return cache_root
 
-        dtype_mapping = {
-            torch.float32: Type.f32,
-            torch.float64: Type.f64,
-            torch.float16: Type.f16,
-            torch.int64: Type.i64,
-            torch.int32: Type.i32,
-            torch.uint8: Type.u8,
-            torch.int8: Type.i8,
-            torch.bool: Type.boolean
-        }
+def cached_model_name(model_hash_str, device, args, cache_root, reversed = False):
+    if model_hash_str is None:
+        return None
 
-        for idx, input_data in enumerate(example_inputs):
-            om.inputs[idx].get_node().set_element_type(dtype_mapping[input_data.dtype])
-            om.inputs[idx].get_node().set_partial_shape(PartialShape(list(input_data.shape)))
-        om.validate_nodes_and_infer_types()
+    model_cache_dir = cache_root + "/model/"
 
-        if model_hash_str is not None:
-            core.set_property({'CACHE_DIR': cache_root + '/blob'})
+    try:
+        os.makedirs(model_cache_dir, exist_ok=True)
+        file_name = model_cache_dir + model_hash_str + "_" + device
+    except OSError as error:
+        shared.log.error(f"Cache directory {cache_root} cannot be created. Model caching is disabled. Error: {error}")
+        return None
 
-        compiled_model = core.compile_model(om, device)
-        def _call(*args):
-            ov_inputs = [a.detach().cpu().numpy() for a in args]
-            ov_inputs.reverse()
-            res = compiled_model(ov_inputs)
-            result = [torch.from_numpy(res[out]) for out in compiled_model.outputs]
-            return result
-        return _call
-    else:
-        example_inputs.reverse()
-        model = make_fx(subgraph)(*example_inputs)
-        with devices.inference_context():
-            model.eval()
-        partitioner = Partitioner()
-        compiled_model = partitioner.make_partitions(model)
+    inputs_str = ""
+    for input_data in args:
+        if reversed:
+            inputs_str = "_" + str(input_data.type()) + str(input_data.size())[11:-1].replace(" ", "") + inputs_str
+        else:
+            inputs_str += "_" + str(input_data.type()) + str(input_data.size())[11:-1].replace(" ", "")
+    inputs_str = sha256(inputs_str.encode('utf-8')).hexdigest()
+    file_name += inputs_str
 
-        def _call(*args):
-            res = execute(compiled_model, *args, executor="openvino",
-                            executor_parameters=executor_parameters)
-            return res
-        return _call
-
-
-def get_cached_file_name(*args, model_hash_str, device, cache_root):
-    file_name = None
-    if model_hash_str is not None:
-        model_cache_dir = cache_root + "/model/"
-        try:
-            os.makedirs(model_cache_dir, exist_ok=True)
-            file_name = model_cache_dir + model_hash_str + "_" + device
-            for input_data in args:
-                if file_name is not None:
-                    file_name += "_" + str(input_data.type()) + str(input_data.size())[11:-1].replace(" ", "")
-        except OSError as error:
-            print("Cache directory ", cache_root, " cannot be created. Model caching is disabled. Error: ", error)
-            file_name = None
-            model_hash_str = None
     return file_name
 
+def check_fully_supported(self, graph_module):
+    num_fused = 0
+    for node in graph_module.graph.nodes:
+        if node.op == "call_module" and "fused_" in node.name:
+            num_fused += 1
+        elif node.op != "placeholder" and node.op != "output":
+            return False
+    if num_fused == 1:
+        return True
+    return False
+
+Partitioner.check_fully_supported = functools.partial(check_fully_supported, Partitioner)
+
+def execute(
+    gm,
+    *args,
+    executor = "openvino",
+    executor_parameters = None,
+    file_name = ""
+):
+    if executor == "openvino":
+        return openvino_execute_partitioned(gm, *args, executor_parameters=executor_parameters, file_name=file_name)
+    elif executor == "strictly_openvino":
+        return openvino_execute(gm, *args, executor_parameters=executor_parameters, file_name=file_name)
+
+    msg = "Received unexpected value for 'executor': {0}. Allowed values are: openvino, strictly_openvino.".format(executor)
+    raise ValueError(msg)
+
+def execute_cached(compiled_model, *args):
+    flat_args, _ = tree_flatten(args)
+    ov_inputs = [a.detach().cpu().numpy() for a in flat_args]
+
+    if (shared.compiled_model_state.cn_model == []):
+        ov_inputs.reverse()
+
+    res = compiled_model(ov_inputs)
+    result = [torch.from_numpy(res[out]) for out in compiled_model.outputs]
+    return result
+
 def openvino_clear_caches():
     global partitioned_modules
     global compiled_cache
 
     compiled_cache.clear()
     partitioned_modules.clear()
+
+def openvino_compile(gm, *args, model_hash_str: str = None, file_name=""):
+    core = Core()
+
+    device = get_device()
+    cache_root = cache_root_path()
+
+    if file_name is not None and os.path.isfile(file_name + ".xml") and os.path.isfile(file_name + ".bin"):
+        om = core.read_model(file_name + ".xml")
+    else:
+        fe_manager = FrontEndManager()
+        fe = fe_manager.load_by_framework("pytorch")
+
+        input_shapes = []
+        input_types = []
+        for input_data in args:
+            input_types.append(input_data.type())
+            input_shapes.append(input_data.size())
+
+        decoder = TorchFXPythonDecoder(gm, gm, input_shapes=input_shapes, input_types=input_types)
+
+        im = fe.load(decoder)
+
+        om = fe.convert(im)
+
+        if (file_name is not None):
+            serialize(om, file_name + ".xml", file_name + ".bin")
+            if (shared.compiled_model_state.cn_model != []):
+                f = open(file_name + ".txt", "w")
+                for input_data in args:
+                    f.write(str(input_data.size()))
+                    f.write("\n")
+                f.close()
+
+    dtype_mapping = {
+        torch.float32: Type.f32,
+        torch.float64: Type.f64,
+        torch.float16: Type.f16,
+        torch.int64: Type.i64,
+        torch.int32: Type.i32,
+        torch.uint8: Type.u8,
+        torch.int8: Type.i8,
+        torch.bool: Type.boolean
+    }
+
+    for idx, input_data in enumerate(args):
+        om.inputs[idx].get_node().set_element_type(dtype_mapping[input_data.dtype])
+        om.inputs[idx].get_node().set_partial_shape(PartialShape(list(input_data.shape)))
+    om.validate_nodes_and_infer_types()
+
+    if model_hash_str is not None:
+        core.set_property({'CACHE_DIR': cache_root + '/blob'})
+
+    compiled = core.compile_model(om, device)
+    return compiled
+
+def openvino_compile_cached_model(cached_model_path, *example_inputs):
+    core = Core()
+    om = core.read_model(cached_model_path + ".xml")
+
+    dtype_mapping = {
+        torch.float32: Type.f32,
+        torch.float64: Type.f64,
+        torch.float16: Type.f16,
+        torch.int64: Type.i64,
+        torch.int32: Type.i32,
+        torch.uint8: Type.u8,
+        torch.int8: Type.i8,
+        torch.bool: Type.boolean
+    }
+
+    for idx, input_data in enumerate(example_inputs):
+        om.inputs[idx].get_node().set_element_type(dtype_mapping[input_data.dtype])
+        om.inputs[idx].get_node().set_partial_shape(PartialShape(list(input_data.shape)))
+    om.validate_nodes_and_infer_types()
+
+    core.set_property({'CACHE_DIR': cache_root_path() + '/blob'})
+
+    compiled_model = core.compile_model(om, get_device())
+
+    return compiled_model
+
+def openvino_execute(gm, *args, executor_parameters=None, partition_id, file_name=""):
+    executor_parameters = executor_parameters or DEFAULT_OPENVINO_PYTHON_CONFIG
+
+    use_cache = executor_parameters.get(
+        "use_python_fusion_cache",
+        DEFAULT_OPENVINO_PYTHON_CONFIG["use_python_fusion_cache"],
+    )
+    global compiled_cache
+
+    model_hash_str = executor_parameters.get("model_hash_str", None)
+    if model_hash_str is not None:
+        model_hash_str = model_hash_str + str(partition_id)
+
+    if use_cache and (partition_id in compiled_cache):
+        compiled = compiled_cache[partition_id]
+    else:
+        if (shared.compiled_model_state.cn_model != [] and file_name is not None
+                and os.path.isfile(file_name + ".xml") and os.path.isfile(file_name + ".bin")):
+            compiled = openvino_compile_cached_model(file_name, *args)
+        else:
+            compiled = openvino_compile(gm, *args, model_hash_str=model_hash_str, file_name=file_name)
+        compiled_cache[partition_id] = compiled
+
+    flat_args, _ = tree_flatten(args)
+    ov_inputs = [a.detach().cpu().numpy() for a in flat_args]
+
+    res = compiled(ov_inputs)
+
+    results1 = [torch.from_numpy(res[out]) for out in compiled.outputs]
+    if len(results1) == 1:
+        return results1[0]
+    return results1
+
+def openvino_execute_partitioned(gm, *args, executor_parameters=None, file_name=""):
+    executor_parameters = executor_parameters or DEFAULT_OPENVINO_PYTHON_CONFIG
+
+    global partitioned_modules
+
+    use_python_fusion_cache = executor_parameters.get(
+        "use_python_fusion_cache",
+        DEFAULT_OPENVINO_PYTHON_CONFIG["use_python_fusion_cache"],
+    )
+    model_hash_str = executor_parameters.get("model_hash_str", None)
+
+    signature = str(id(gm))
+    for idx, input_data in enumerate(args):
+        if isinstance(input_data, torch.Tensor):
+            signature = signature + "_" + str(idx) + ":" + str(input_data.type())[6:] + ":" + str(input_data.size())[11:-1].replace(" ", "")
+        else:
+            signature = signature + "_" + str(idx) + ":" + type(input_data).__name__ + ":val(" + str(input_data) + ")"
+
+    if signature not in partitioned_modules:
+        partitioned_modules[signature] = partition_graph(gm, use_python_fusion_cache=use_python_fusion_cache,
+                                                         model_hash_str=model_hash_str, file_name=file_name)
+
+    return partitioned_modules[signature](*args)
+
+def partition_graph(gm, use_python_fusion_cache: bool, model_hash_str: str = None, file_name=""):
+    global max_openvino_partitions
+    for node in gm.graph.nodes:
+        if node.op == "call_module" and "fused_" in node.name:
+            openvino_submodule = getattr(gm, node.name)
+            gm.delete_submodule(node.target)
+            gm.add_submodule(
+                node.target,
+                OpenVINOGraphModule(openvino_submodule, shared.compiled_model_state.partition_id, use_python_fusion_cache,
+                        model_hash_str=model_hash_str, file_name=file_name),
+            )
+            shared.compiled_model_state.partition_id = shared.compiled_model_state.partition_id + 1
+
+    return gm
+
+@register_backend
+@fake_tensor_unsupported
+def openvino_fx(subgraph, example_inputs):
+    executor_parameters = None
+    inputs_reversed = False
+    if os.getenv("OPENVINO_TORCH_MODEL_CACHING") != "0":
+        os.environ.setdefault('OPENVINO_TORCH_MODEL_CACHING', "1")
+        # Create a hash to be used for caching
+        model_hash_str = sha256(subgraph.code.encode('utf-8')).hexdigest()
+        if (shared.compiled_model_state.cn_model != [] and shared.compiled_model_state.partition_id == 0):
+            model_hash_str = model_hash_str + str(shared.compiled_model_state.cn_model)
+
+        if (shared.compiled_model_state.lora_model != []):
+            model_hash_str = model_hash_str + str(shared.compiled_model_state.lora_model)
+
+        executor_parameters = {"model_hash_str": model_hash_str}
+        # Check if the model was fully supported and already cached
+        example_inputs.reverse()
+        inputs_reversed = True
+        maybe_fs_cached_name = cached_model_name(model_hash_str + "_fs", get_device(), example_inputs, cache_root_path())
+
+        if os.path.isfile(maybe_fs_cached_name + ".xml") and os.path.isfile(maybe_fs_cached_name + ".bin"):
+            if (shared.compiled_model_state.cn_model != [] and str(shared.compiled_model_state.cn_model) in maybe_fs_cached_name):
+                example_inputs_reordered = []
+                if (os.path.isfile(maybe_fs_cached_name + ".txt")):
+                    f = open(maybe_fs_cached_name + ".txt", "r")
+                    for input_data in example_inputs:
+                        shape = f.readline()
+                        if (str(input_data.size()) != shape):
+                            for idx1, input_data1 in enumerate(example_inputs):
+                                if (str(input_data1.size()).strip() == str(shape).strip()):
+                                    example_inputs_reordered.append(example_inputs[idx1])
+                    example_inputs = example_inputs_reordered
+
+                # Model is fully supported and already cached. Run the cached OV model directly.
+                compiled_model = openvino_compile_cached_model(maybe_fs_cached_name, *example_inputs)
+
+                def _call(*args):
+                    if (shared.compiled_model_state.cn_model != [] and str(shared.compiled_model_state.cn_model) in maybe_fs_cached_name):
+                        args_reordered = []
+                        if (os.path.isfile(maybe_fs_cached_name + ".txt")):
+                            f = open(maybe_fs_cached_name + ".txt", "r")
+                            for input_data in args:
+                                shape = f.readline()
+                                if (str(input_data.size()) != shape):
+                                    for idx1, input_data1 in enumerate(args):
+                                        if (str(input_data1.size()).strip() == str(shape).strip()):
+                                            args_reordered.append(args[idx1])
+                        args = args_reordered
+
+                    res = execute_cached(compiled_model, *args)
+                    shared.compiled_model_state.partition_id = shared.compiled_model_state.partition_id + 1
+                    return res
+                return _call
+    else:
+        maybe_fs_cached_name = ""
+
+    if inputs_reversed:
+        example_inputs.reverse()
+    model = make_fx(subgraph)(*example_inputs)
+    for node in model.graph.nodes:
+        if node.target == torch.ops.aten.mul_.Tensor:
+            node.target = torch.ops.aten.mul.Tensor
+    with devices.inference_context():
+        model.eval()
+    partitioner = Partitioner()
+    compiled_model = partitioner.make_partitions(model)
+
+    if executor_parameters is not None and 'model_hash_str' in executor_parameters:
+        # Check if the model is fully supported.
+        fully_supported = partitioner.check_fully_supported(compiled_model)
+        if fully_supported:
+            executor_parameters["model_hash_str"] += "_fs"
+
+    def _call(*args):
+        res = execute(compiled_model, *args, executor="openvino",
+                        executor_parameters=executor_parameters, file_name=maybe_fs_cached_name)
+        return res
+    return _call

modules/lora_diffusers.py

@@ -1,3 +1,4 @@
+import time
 import diffusers
 import diffusers.models.lora as diffusers_lora
 # from modules import shared
@@ -7,13 +8,15 @@ import modules.shared as shared
 lora_state = { # TODO Lora state for Diffusers
     'multiplier': [],
     'active': False,
-    'loaded': 0,
-    'all_loras': []
+    'loaded': [],
+    'all_loras': [],
 }
 def unload_diffusers_lora():
     try:
         pipe = shared.sd_model
-        if shared.opts.diffusers_lora_loader == "diffusers default":
+        if shared.opts.diffusers_lora_loader == "diffusers":
+            if len(lora_state['loaded']) > 1 and hasattr(pipe, "unfuse_lora"):
+                pipe.unfuse_lora()
             pipe.unload_lora_weights()
             pipe._remove_text_encoder_monkey_patch() # pylint: disable=W0212
             proc_cls_name = next(iter(pipe.unet.attn_processors.values())).__class__.__name__
@@ -29,22 +32,32 @@ def unload_diffusers_lora():
                 if shared.opts.diffusers_lora_loader == "sequential apply":
                     lora_network.unapply_to()
         lora_state['active'] = False
-        lora_state['loaded'] = 0
+        lora_state['loaded'].clear()
         lora_state['all_loras'] = []
         lora_state['multiplier'] = []
-
     except Exception as e:
-        shared.log.error(f"Diffusers LoRA unloading failed: {e}")
+        shared.log.error(f"LoRA unload failed: {e}")
 
 
-def load_diffusers_lora(name, lora, strength = 1.0):
+def load_diffusers_lora(name, lora, strength = 1.0, num_loras = 1):
+    if f'{lora.filename}:{strength}' in lora_state['loaded']:
+        shared.log.info(f'LoRA cached: {name} strength={strength}')
+        return
     try:
+        t0 = time.time()
         pipe = shared.sd_model
         lora_state['active'] = True
-        lora_state['loaded'] += 1
         lora_state['multiplier'].append(strength)
-        if shared.opts.diffusers_lora_loader == "diffusers default":
-            pipe.load_lora_weights(lora.filename, cache_dir=shared.opts.diffusers_dir, local_files_only=True, lora_scale=strength)
+        fuse = 0
+        if shared.opts.diffusers_lora_loader.startswith("diffusers"):
+            pipe.load_lora_weights(lora.filename, cache_dir=shared.opts.diffusers_dir, local_files_only=True, lora_scale=strength, low_cpu_mem_usage=True)
+            if num_loras > 1 and hasattr(pipe, "fuse_lora"):
+                t2 = time.time()
+                pipe.fuse_lora(lora_scale=strength)
+                fuse = time.time() - t2
+            lora_state['loaded'].append(f'{lora.filename}:{strength}')
+            if shared.compiled_model_state is not None: #filename breaks caching
+                shared.compiled_model_state.lora_model.append(f'{name}:{strength}')
         else:
             from safetensors.torch import load_file
             lora_sd = load_file(lora.filename)
@@ -60,20 +73,26 @@ def load_diffusers_lora(name, lora, strength = 1.0):
                 lora_network.to(shared.device, dtype=pipe.unet.dtype)
                 lora_network.apply_to(multiplier=strength)
             lora_state['all_loras'].append(lora_network)
-        shared.log.info(f"LoRA loaded: {name} strength={strength} loader={shared.opts.diffusers_lora_loader}")
+            lora_state['loaded'].append(f'{lora.filename}:{strength}')
+            if shared.compiled_model_state is not None: #filename breaks caching
+                shared.compiled_model_state.lora_model.append(f'{name}:{strength}')
+        t1 = time.time()
+        fuse = f'fuse={fuse:.2f}s' if fuse > 0 else ''
+        shared.log.info(f'LoRA loaded: {name} strength={strength} loader="{shared.opts.diffusers_lora_loader}" lora={t1-t0:.2f}s {fuse}')
     except Exception as e:
-        shared.log.error(f"LoRA loading failed: {name} {e}")
+        lines = str(e).splitlines()
+        shared.log.error(f'LoRA loading failed: {name} loader="{shared.opts.diffusers_lora_loader}" {lines[0]}')
 
 
 # Diffusersで動くLoRA。このファイル単独で完結する。
 # LoRA module for Diffusers. This file works independently.
-import bisect
-import math
-from typing import Any, Dict, List, Mapping, Optional, Union
-from diffusers import UNet2DConditionModel
-from tqdm import tqdm
-from transformers import CLIPTextModel
-import torch
+import bisect # pylint: disable=wrong-import-order
+import math # pylint: disable=wrong-import-order
+from typing import Any, Dict, List, Mapping, Optional, Union # pylint: disable=wrong-import-order
+from diffusers import UNet2DConditionModel # pylint: disable=wrong-import-order
+from tqdm import tqdm # pylint: disable=wrong-import-order
+from transformers import CLIPTextModel # pylint: disable=wrong-import-order
+import torch # pylint: disable=wrong-import-order
 
 
 def make_unet_conversion_map() -> Dict[str, str]:
@@ -496,7 +515,7 @@ class LoRANetwork(torch.nn.Module): # pylint: disable=abstract-method
         for lora in tqdm(self.text_encoder_loras + self.unet_loras):
             lora.restore_from(multiplier)
 
-    def load_state_dict(self, state_dict: Mapping[str, Any], strict: bool = True):
+    def load_state_dict(self, state_dict: Mapping[str, Any], strict: bool = True): # pylint: disable=arguments-differ
         # convert SDXL Stability AI's state dict to Diffusers' based state dict
         map_keys = list(UNET_CONVERSION_MAP.keys())  # prefix of U-Net modules
         map_keys.sort()
@@ -514,8 +533,8 @@ class LoRANetwork(torch.nn.Module): # pylint: disable=abstract-method
         # because V2 LoRA is based on U-Net created by use_linear_projection=False
         my_state_dict = self.state_dict()
         for key in state_dict.keys():
-            if state_dict[key].size() != my_state_dict[key].size():
+            if state_dict[key].size() != my_state_dict[key].size(): # pylint: disable=unsubscriptable-object
                 # print(f"convert {key} from {state_dict[key].size()} to {my_state_dict[key].size()}")
-                state_dict[key] = state_dict[key].view(my_state_dict[key].size())
+                state_dict[key] = state_dict[key].view(my_state_dict[key].size()) # pylint: disable=unsubscriptable-object
 
         return super().load_state_dict(state_dict, strict)

modules/paths.py

@@ -68,10 +68,12 @@ def create_paths(opts, log=None):
         fullpath = os.path.join(data_path, tgt)
         if len(data_path) > 0 and os.path.isabs(data_path):
             return fullpath
+        if os.path.isabs(fullpath) and os.path.exists(fullpath):
+            return fullpath
         try:
             relpath = os.path.relpath(fullpath, script_path)
             opts.data[folder] = relpath
-        except:
+        except Exception:
             opts.data[folder] = fullpath
         return opts.data[folder]
 

modules/postprocessing.py

@@ -17,6 +17,7 @@ def run_postprocessing(extras_mode, image, image_folder: List[tempfile.NamedTemp
     outputs = []
     params = {}
     if extras_mode == 1:
+        shared.log.debug(f'process: mode=batch folder={image_folder}')
         for img in image_folder:
             if isinstance(img, Image.Image):
                 image = img
@@ -29,6 +30,7 @@ def run_postprocessing(extras_mode, image, image_folder: List[tempfile.NamedTemp
             image_names.append(fn)
             image_ext.append(ext)
     elif extras_mode == 2:
+        shared.log.debug(f'process: mode=folder folder={input_dir}')
         assert not shared.cmd_opts.hide_ui_dir_config, '--hide-ui-dir-config option must be disabled'
         assert input_dir, 'input directory not selected'
         image_list = shared.listfiles(input_dir)
@@ -50,6 +52,7 @@ def run_postprocessing(extras_mode, image, image_folder: List[tempfile.NamedTemp
     else:
         outpath = opts.outdir_samples or opts.outdir_extras_samples
     for image, name, ext in zip(image_data, image_names, image_ext):
+        shared.log.debug(f'process: image={image} {args}')
         infotext = ''
         if shared.state.interrupted:
             shared.log.debug('Postprocess interrupted')

modules/processing.py

@@ -462,7 +462,8 @@ def create_infotext(p: StableDiffusionProcessing, all_prompts, all_seeds, all_su
     if all_negative_prompts is None:
         all_negative_prompts = p.all_negative_prompts
     comment = ', '.join(comments) if comments is not None and type(comments) is list else None
-
+    ops = list(set(p.ops))
+    ops.reverse()
     args = {
         # basic
         "Steps": p.steps,
@@ -488,7 +489,7 @@ def create_infotext(p: StableDiffusionProcessing, all_prompts, all_seeds, all_su
         "Backend": 'Diffusers' if shared.backend == shared.Backend.DIFFUSERS else 'Original',
         "Version": git_commit,
         "Comment": comment,
-        "Operations": '; '.join(p.ops).replace('"', '') if len(p.ops) > 0 else 'none',
+        "Operations": '; '.join(ops).replace('"', '') if len(p.ops) > 0 else 'none',
     }
     if 'txt2img' in p.ops:
         pass
@@ -572,6 +573,8 @@ def print_profile(profile, msg: str):
 def process_images(p: StableDiffusionProcessing) -> Processed:
     if not hasattr(p.sd_model, 'sd_checkpoint_info'):
         return None
+    if p.scripts is not None:
+        p.scripts.before_process(p)
     stored_opts = {}
     for k, v in p.override_settings.copy().items():
         orig = shared.opts.data.get(k, None) or shared.opts.data_labels[k].default
@@ -686,8 +689,8 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
         p.all_subseeds = subseed
     else:
         p.all_subseeds = [int(subseed) + x for x in range(len(p.all_prompts))]
-    if os.path.exists(shared.opts.embeddings_dir) and not p.do_not_reload_embeddings:
-        modules.sd_hijack.model_hijack.embedding_db.load_textual_inversion_embeddings()
+    if os.path.exists(shared.opts.embeddings_dir) and not p.do_not_reload_embeddings and shared.backend == shared.Backend.ORIGINAL:
+        modules.sd_hijack.model_hijack.embedding_db.load_textual_inversion_embeddings(force_reload=False)
     if p.scripts is not None:
         p.scripts.process(p)
     infotexts = []
@@ -984,12 +987,14 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing):
             info = create_infotext(self, self.all_prompts, self.all_seeds, self.all_subseeds, [], iteration=self.iteration, position_in_batch=index)
             self.extra_generation_params = orig1
             self.restore_faces = orig2
-            images.save_image(image, self.outpath_samples, "", seeds[index], prompts[index], shared.opts.samples_format, info=info, suffix="-before-highres-fix")
+            images.save_image(image, self.outpath_samples, "", seeds[index], prompts[index], shared.opts.samples_format, info=info, suffix="-before-hires", index=index+1)
 
         if shared.backend == shared.Backend.DIFFUSERS:
             modules.sd_models.set_diffuser_pipe(self.sd_model, modules.sd_models.DiffusersTaskType.TEXT_2_IMAGE)
 
         latent_scale_mode = shared.latent_upscale_modes.get(self.hr_upscaler, None) if self.hr_upscaler is not None else shared.latent_upscale_modes.get(shared.latent_upscale_default_mode, "None")
+        if latent_scale_mode is not None:
+            self.hr_force = False # no need to force anything
         if self.enable_hr and (latent_scale_mode is None or self.hr_force):
             if len([x for x in shared.sd_upscalers if x.name == self.hr_upscaler]) == 0:
                 shared.log.warning(f"Cannot find upscaler for hires: {self.hr_upscaler}")
@@ -1013,11 +1018,10 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing):
                 decoded_samples = decode_first_stage(self.sd_model, samples.to(dtype=devices.dtype_vae))
                 lowres_samples = torch.clamp((decoded_samples + 1.0) / 2.0, min=0.0, max=1.0)
                 batch_images = []
-                for i, x_sample in enumerate(lowres_samples):
+                for _i, x_sample in enumerate(lowres_samples):
                     x_sample = 255. * np.moveaxis(x_sample.cpu().numpy(), 0, 2)
                     x_sample = validate_sample(x_sample)
                     image = Image.fromarray(x_sample)
-                    save_intermediate(image, i)
                     image = images.resize_image(1, image, target_width, target_height, upscaler_name=self.hr_upscaler)
                     image = np.array(image).astype(np.float32) / 255.0
                     image = np.moveaxis(image, 2, 0)

modules/processing_diffusers.py

@@ -1,8 +1,10 @@
 import time
+import math
 import inspect
 import typing
 import torch
 import torchvision.transforms.functional as TF
+import diffusers
 import modules.devices as devices
 import modules.shared as shared
 import modules.sd_samplers as sd_samplers
@@ -15,12 +17,6 @@ from modules.processing import StableDiffusionProcessing
 import modules.prompt_parser_diffusers as prompt_parser_diffusers
 
 
-try:
-    import diffusers
-except Exception as ex:
-    shared.log.error(f'Failed to import diffusers: {ex}')
-
-
 def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_prompts):
     results = []
     if p.enable_hr and p.hr_upscaler != 'None' and p.denoising_strength > 0 and len(getattr(p, 'init_images', [])) == 0:
@@ -78,8 +74,8 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
             model.vae.to(devices.device)
         latents.to(model.vae.device)
 
-        needs_upcasting = model.vae.dtype == torch.float16 and model.vae.config.force_upcast
-        if needs_upcasting: # this is done by diffusers automatically if output_type != 'latent'
+        upcast = (model.vae.dtype == torch.float16) and model.vae.config.force_upcast and hasattr(model, 'upcast_vae')
+        if upcast: # this is done by diffusers automatically if output_type != 'latent'
             model.upcast_vae()
             latents = latents.to(next(iter(model.vae.post_quant_conv.parameters())).dtype)
 
@@ -87,7 +83,7 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
         if shared.opts.diffusers_move_unet and not model.has_accelerate:
             model.unet.to(unet_device)
         t1 = time.time()
-        shared.log.debug(f'VAE decode: name={sd_vae.loaded_vae_file if sd_vae.loaded_vae_file is not None else "baked"} dtype={model.vae.dtype} upcast={model.vae.config.get("force_upcast", None)} images={latents.shape[0]} latents={latents.shape} time={round(t1-t0, 3)}s')
+        shared.log.debug(f'VAE decode: name={sd_vae.loaded_vae_file if sd_vae.loaded_vae_file is not None else "baked"} dtype={model.vae.dtype} upcast={upcast} images={latents.shape[0]} latents={latents.shape} time={round(t1-t0, 3)}s')
         return decoded
 
     def full_vae_encode(image, model):
@@ -171,6 +167,8 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
         return prompts, negative_prompts, prompts_2, negative_prompts_2
 
     def set_pipeline_args(model, prompts: list, negative_prompts: list, prompts_2: typing.Optional[list]=None, negative_prompts_2: typing.Optional[list]=None, desc:str='', **kwargs):
+        if hasattr(model, 'embedding_db'):
+            del model.embedding_db
         try:
             is_refiner = model.text_encoder.__class__.__name__ != 'CLIPTextModel'
         except Exception:
@@ -187,8 +185,13 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
         negative_embed = None
         negative_pooled = None
         prompts, negative_prompts, prompts_2, negative_prompts_2 = fix_prompts(prompts, negative_prompts, prompts_2, negative_prompts_2)
-        if shared.opts.prompt_attention in {'Compel parser', 'Full parser'} and 'StableDiffusion' in model.__class__.__name__:
-            prompt_embed, pooled, negative_embed, negative_pooled = prompt_parser_diffusers.compel_encode_prompts(model, prompts, negative_prompts, prompts_2, negative_prompts_2, is_refiner, kwargs.pop("clip_skip", None))
+        parser = 'Fixed attention'
+        if shared.opts.prompt_attention != 'Fixed attention' and 'StableDiffusion' in model.__class__.__name__:
+            try:
+                prompt_embed, pooled, negative_embed, negative_pooled = prompt_parser_diffusers.compel_encode_prompts(model, prompts, negative_prompts, prompts_2, negative_prompts_2, is_refiner, kwargs.pop("clip_skip", None))
+                parser = shared.opts.prompt_attention
+            except Exception as e:
+                shared.log.error(f'Prompt parser: {e}')
         if 'prompt' in possible:
             if hasattr(model, 'text_encoder') and 'prompt_embeds' in possible and prompt_embed is not None:
                 if type(pooled) == list:
@@ -244,7 +247,10 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
         if 'negative_pooled_prompt_embeds' in clean:
             clean['negative_pooled_prompt_embeds'] = clean['negative_pooled_prompt_embeds'].shape if torch.is_tensor(clean['negative_pooled_prompt_embeds']) else type(clean['negative_pooled_prompt_embeds'])
         clean['generator'] = generator_device
+        clean['parser'] = parser
         shared.log.debug(f'Diffuser pipeline: {model.__class__.__name__} task={sd_models.get_diffusers_task(model)} set={clean}')
+        # components = [{ k: getattr(v, 'device', None) } for k, v in model.components.items()]
+        # shared.log.debug(f'Diffuser pipeline components: {components}')
         return args
 
     def recompile_model(hires=False):
@@ -252,10 +258,12 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
             if shared.opts.cuda_compile_backend == "openvino_fx":
                 compile_height = p.height if not hires else p.hr_upscale_to_y
                 compile_width = p.width if not hires else p.hr_upscale_to_x
-                if (not hasattr(shared.sd_model, "compiled_model_state") or (not shared.sd_model.compiled_model_state.first_pass
-                and (shared.sd_model.compiled_model_state.height != compile_height or shared.sd_model.compiled_model_state.width != compile_width
-                or shared.sd_model.compiled_model_state.batch_size != p.batch_size))):
-                    shared.log.info("OpenVINO: Resolution change detected")
+                if (shared.compiled_model_state is None or
+                (not shared.compiled_model_state.first_pass
+                and (shared.compiled_model_state.height != compile_height
+                or shared.compiled_model_state.width != compile_width
+                or shared.compiled_model_state.batch_size != p.batch_size))):
+                    shared.log.info("OpenVINO: Parameter change detected")
                     shared.log.info("OpenVINO: Recompiling base model")
                     sd_models.unload_model_weights(op='model')
                     sd_models.reload_model_weights(op='model')
@@ -263,19 +271,20 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
                         shared.log.info("OpenVINO: Recompiling refiner")
                         sd_models.unload_model_weights(op='refiner')
                         sd_models.reload_model_weights(op='refiner')
-                shared.sd_model.compiled_model_state.height = compile_height
-                shared.sd_model.compiled_model_state.width = compile_width
-                shared.sd_model.compiled_model_state.batch_size = p.batch_size
-                shared.sd_model.compiled_model_state.first_pass = False
+                shared.compiled_model_state.height = compile_height
+                shared.compiled_model_state.width = compile_width
+                shared.compiled_model_state.batch_size = p.batch_size
+                shared.compiled_model_state.first_pass = False
             else:
                 pass #Can be implemented for TensorRT or Olive
         else:
             pass #Do nothing if compile is disabled
 
+    recompile_model()
+
     is_karras_compatible = shared.sd_model.__class__.__init__.__annotations__.get("scheduler", None) == diffusers.schedulers.scheduling_utils.KarrasDiffusionSchedulers
-    use_sampler = p.sampler_name if not p.is_hr_pass else p.latent_sampler
-    if (not hasattr(shared.sd_model.scheduler, 'name')) or (shared.sd_model.scheduler.name != use_sampler) and (use_sampler != 'Default') and is_karras_compatible:
-        sampler = sd_samplers.all_samplers_map.get(use_sampler, None)
+    if ((not hasattr(shared.sd_model.scheduler, 'name')) or (p.sampler_name == 'DPM SDE') or (shared.sd_model.scheduler.name != p.sampler_name)) and (p.sampler_name != 'Default') and is_karras_compatible:
+        sampler = sd_samplers.all_samplers_map.get(p.sampler_name, None)
         if sampler is None:
             sampler = sd_samplers.all_samplers_map.get("UniPC")
         sd_samplers.create_sampler(sampler.name, shared.sd_model) # TODO(Patrick): For wrapped pipelines this is currently a no-op
@@ -297,20 +306,18 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
     task_specific_kwargs={}
     if sd_models.get_diffusers_task(shared.sd_model) == sd_models.DiffusersTaskType.TEXT_2_IMAGE:
         p.ops.append('txt2img')
-        task_specific_kwargs = {"height": p.height, "width": p.width}
+        task_specific_kwargs = {"height": 8 * math.ceil(p.height / 8), "width": 8 * math.ceil(p.width / 8)}
     elif sd_models.get_diffusers_task(shared.sd_model) == sd_models.DiffusersTaskType.IMAGE_2_IMAGE:
         p.ops.append('img2img')
         task_specific_kwargs = {"image": p.init_images, "strength": p.denoising_strength}
     elif sd_models.get_diffusers_task(shared.sd_model) == sd_models.DiffusersTaskType.INPAINTING:
         p.ops.append('inpaint')
-        task_specific_kwargs = {"image": p.init_images, "mask_image": p.mask, "strength": p.denoising_strength, "height": p.height, "width": p.width}
+        task_specific_kwargs = {"image": p.init_images, "mask_image": p.mask, "strength": p.denoising_strength, "height": 8 * math.ceil(p.height / 8), "width": 8 * math.ceil(p.width / 8)}
 
     if shared.state.interrupted or shared.state.skipped:
         unload_diffusers_lora()
         return results
 
-    recompile_model()
-
     if shared.opts.diffusers_move_base and not shared.sd_model.has_accelerate:
         shared.sd_model.to(devices.device)
 
@@ -373,6 +380,11 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
             if latent_scale_mode is not None or p.hr_force:
                 p.ops.append('hires')
                 recompile_model(hires=True)
+                if ((not hasattr(shared.sd_model.scheduler, 'name')) or (p.latent_sampler == 'DPM SDE') or (shared.sd_model.scheduler.name != p.latent_sampler)) and (p.latent_sampler != 'Default') and is_karras_compatible:
+                    sampler = sd_samplers.all_samplers_map.get(p.latent_sampler, None)
+                    if sampler is None:
+                        sampler = sd_samplers.all_samplers_map.get("UniPC")
+                    sd_samplers.create_sampler(sampler.name, shared.sd_model) # TODO(Patrick): For wrapped pipelines this is currently a no-op
                 sd_models.set_diffuser_pipe(shared.sd_model, sd_models.DiffusersTaskType.IMAGE_2_IMAGE)
                 hires_args = set_pipeline_args(
                     model=shared.sd_model,
@@ -404,7 +416,7 @@ def process_diffusers(p: StableDiffusionProcessing, seeds, prompts, negative_pro
             shared.sd_model.to(devices.cpu)
             devices.torch_gc()
 
-        if (not hasattr(shared.sd_refiner.scheduler, 'name')) or (shared.sd_refiner.scheduler.name != p.latent_sampler) and (p.sampler_name != 'Default'):
+        if ((not hasattr(shared.sd_refiner.scheduler, 'name')) or (p.latent_sampler == 'DPM SDE') or (shared.sd_refiner.scheduler.name != p.latent_sampler)) and (p.latent_sampler != 'Default'):
             sampler = sd_samplers.all_samplers_map.get(p.latent_sampler, None)
             if sampler is None:
                 sampler = sd_samplers.all_samplers_map.get("UniPC")

modules/prompt_parser_diffusers.py

@@ -179,6 +179,7 @@ def compel_encode_prompt(
         return prompt_embed, positive_pooled, negative_embed, negative_pooled
 
     # neither base+sdxl nor refiner+sdxl
-    positive, negative = compel_te1(prompt), compel_te1(negative_prompt)
+    positive = compel_te1(prompt)
+    negative = compel_te1(negative_prompt)
     [prompt_embed, negative_embed] = compel_te1.pad_conditioning_tensors_to_same_length([positive, negative])
     return prompt_embed, None, negative_embed, None

modules/realesrgan_model.py

@@ -1,12 +1,9 @@
 import os
-import sys
 import numpy as np
 from PIL import Image
-from basicsr.utils.download_util import load_file_from_url
 from modules.upscaler import Upscaler, UpscalerData
-from modules.shared import opts, device
+from modules.shared import opts, device, log
 from modules import modelloader
-import modules.errors as errors
 
 
 class UpscalerRealESRGAN(Upscaler):
@@ -16,8 +13,7 @@ class UpscalerRealESRGAN(Upscaler):
         super().__init__()
         try:
             from basicsr.archs.rrdbnet_arch import RRDBNet # pylint: disable=unused-import
-            from realesrgan import RealESRGANer # pylint: disable=unused-import
-            from realesrgan.archs.srvgg_arch import SRVGGNetCompact # pylint: disable=unused-import
+            from modules.realesrgan_model_arch import RealESRGANer, SRVGGNetCompact # pylint: disable=unused-import
             self.enable = True
             self.scalers = []
             scalers = self.load_models(path)
@@ -28,11 +24,9 @@ class UpscalerRealESRGAN(Upscaler):
                     local_model_candidates = [local_model for local_model in local_model_paths if local_model.endswith(f"{filename}.pth")]
                     if local_model_candidates:
                         scaler.local_data_path = local_model_candidates[0]
-                if scaler.name in opts.realesrgan_enabled_models:
-                    self.scalers.append(scaler)
-
+                self.scalers.append(scaler)
         except Exception as e:
-            errors.display(e, 'real-esrgan')
+            log.error(f"Error loading Real-ESRGAN: model={path} {e}")
             self.enable = False
             self.scalers = []
 
@@ -41,14 +35,13 @@ class UpscalerRealESRGAN(Upscaler):
             return img
 
         try:
-            from realesrgan import RealESRGANer
+            from modules.realesrgan_model_arch import RealESRGANer
         except Exception:
-            print("Error importing Real-ESRGAN:", file=sys.stderr)
+            log.error("Error importing Real-ESRGAN:")
             return img
 
         info = self.load_model(selected_model)
-        if not os.path.exists(info.local_data_path):
-            print(f"Unable to load RealESRGAN model: {info.name}")
+        if info is None or not os.path.exists(info.local_data_path):
             return img
 
         upsampler = RealESRGANer(
@@ -70,13 +63,15 @@ class UpscalerRealESRGAN(Upscaler):
         try:
             info = next(iter([scaler for scaler in self.scalers if scaler.data_path == path]), None)
             if info is None:
-                print(f"Unable to find model info: {path}")
+                log.error(f"Model failed loading: type=R-ESRGAN model={info.name}")
                 return None
             if info.local_data_path.startswith("http"):
+                from modules.modelloader import load_file_from_url
                 info.local_data_path = load_file_from_url(url=info.data_path, model_dir=self.model_download_path, progress=True)
+            log.info(f"Model loaded: type=R-ESRGAN model={info.name}")
             return info
         except Exception as e:
-            errors.display(e, 'real-esrgan model list')
+            log.error(f"Model failed loading: type=R-ESRGAN model={info.name} {e}")
         return None
 
     def load_models(self, _):
@@ -86,7 +81,7 @@ class UpscalerRealESRGAN(Upscaler):
 def get_realesrgan_models(scaler):
     try:
         from basicsr.archs.rrdbnet_arch import RRDBNet
-        from realesrgan.archs.srvgg_arch import SRVGGNetCompact
+        from modules.realesrgan_model_arch import SRVGGNetCompact # pylint: disable=unused-import
         models = [
             UpscalerData(
                 name="R-ESRGAN General 4xV3",
@@ -132,6 +127,6 @@ def get_realesrgan_models(scaler):
             ),
         ]
         return models
-    except Exception:
-        print("Error creating Real-ESRGAN models list", file=sys.stderr)
+    except Exception as e:
+        log.error(f'Error creating Real-ESRGAN models list: {e}')
         return []

modules/realesrgan_model_arch.py

@@ -0,0 +1,384 @@
+import os
+import math
+import queue
+import threading
+import cv2
+import numpy as np
+import torch
+from torch.nn import functional as F
+from rich.progress import Progress, TextColumn, BarColumn, TaskProgressColumn, TimeRemainingColumn, TimeElapsedColumn
+from modules.shared import log, console
+
+ROOT_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+
+
+class RealESRGANer():
+    """A helper class for upsampling images with RealESRGAN.
+
+    Args:
+        scale (int): Upsampling scale factor used in the networks. It is usually 2 or 4.
+        model_path (str): The path to the pretrained model. It can be urls (will first download it automatically).
+        model (nn.Module): The defined network. Default: None.
+        tile (int): As too large images result in the out of GPU memory issue, so this tile option will first crop
+            input images into tiles, and then process each of them. Finally, they will be merged into one image.
+            0 denotes for do not use tile. Default: 0.
+        tile_pad (int): The pad size for each tile, to remove border artifacts. Default: 10.
+        pre_pad (int): Pad the input images to avoid border artifacts. Default: 10.
+        half (float): Whether to use half precision during inference. Default: False.
+    """
+
+    def __init__(self,
+                 scale,
+                 model_path,
+                 dni_weight=None,
+                 model=None,
+                 tile=0,
+                 tile_pad=10,
+                 pre_pad=10,
+                 half=False,
+                 device=None,
+                 gpu_id=None):
+        self.scale = scale
+        self.tile_size = tile
+        self.tile_pad = tile_pad
+        self.pre_pad = pre_pad
+        self.mod_scale = None
+        self.half = half
+
+        # initialize model
+        if gpu_id:
+            self.device = torch.device(
+                f'cuda:{gpu_id}' if torch.cuda.is_available() else 'cpu') if device is None else device
+        else:
+            self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') if device is None else device
+
+        if isinstance(model_path, list):
+            # dni
+            assert len(model_path) == len(dni_weight), 'model_path and dni_weight should have the save length.'
+            loadnet = self.dni(model_path[0], model_path[1], dni_weight)
+        else:
+            # if the model_path starts with https, it will first download models to the folder: weights
+            if model_path.startswith('https://'):
+                from modules.modelloader import load_file_from_url
+                model_path = load_file_from_url(url=model_path, model_dir=os.path.join(ROOT_DIR, 'weights'), progress=True, file_name=None)
+            loadnet = torch.load(model_path, map_location=torch.device('cpu'))
+
+        # prefer to use params_ema
+        if 'params_ema' in loadnet:
+            keyname = 'params_ema'
+        else:
+            keyname = 'params'
+        model.load_state_dict(loadnet[keyname], strict=True)
+
+        model.eval()
+        self.model = model.to(self.device)
+        if self.half:
+            self.model = self.model.half()
+
+    def dni(self, net_a, net_b, dni_weight, key='params', loc='cpu'):
+        """Deep network interpolation.
+
+        ``Paper: Deep Network Interpolation for Continuous Imagery Effect Transition``
+        """
+        net_a = torch.load(net_a, map_location=torch.device(loc))
+        net_b = torch.load(net_b, map_location=torch.device(loc))
+        for k, v_a in net_a[key].items():
+            net_a[key][k] = dni_weight[0] * v_a + dni_weight[1] * net_b[key][k]
+        return net_a
+
+    def pre_process(self, img):
+        """Pre-process, such as pre-pad and mod pad, so that the images can be divisible
+        """
+        img = torch.from_numpy(np.transpose(img, (2, 0, 1))).float()
+        self.img = img.unsqueeze(0).to(self.device)
+        if self.half:
+            self.img = self.img.half()
+
+        # pre_pad
+        if self.pre_pad != 0:
+            self.img = F.pad(self.img, (0, self.pre_pad, 0, self.pre_pad), 'reflect')
+        # mod pad for divisible borders
+        if self.scale == 2:
+            self.mod_scale = 2
+        elif self.scale == 1:
+            self.mod_scale = 4
+        if self.mod_scale is not None:
+            self.mod_pad_h, self.mod_pad_w = 0, 0
+            _, _, h, w = self.img.size()
+            if (h % self.mod_scale != 0):
+                self.mod_pad_h = (self.mod_scale - h % self.mod_scale)
+            if (w % self.mod_scale != 0):
+                self.mod_pad_w = (self.mod_scale - w % self.mod_scale)
+            self.img = F.pad(self.img, (0, self.mod_pad_w, 0, self.mod_pad_h), 'reflect')
+
+    def process(self):
+        # model inference
+        self.output = self.model(self.img)
+
+    def tile_process(self):
+        """It will first crop input images to tiles, and then process each tile.
+        Finally, all the processed tiles are merged into one images.
+
+        Modified from: https://github.com/ata4/esrgan-launcher
+        """
+        batch, channel, height, width = self.img.shape
+        output_height = height * self.scale
+        output_width = width * self.scale
+        output_shape = (batch, channel, output_height, output_width)
+
+        # start with black image
+        self.output = self.img.new_zeros(output_shape)
+        tiles_x = math.ceil(width / self.tile_size)
+        tiles_y = math.ceil(height / self.tile_size)
+
+        # loop over all tiles
+        with Progress(TextColumn('[cyan]{task.description}'), BarColumn(), TaskProgressColumn(), TimeRemainingColumn(), TimeElapsedColumn(), console=console) as progress:
+            task = progress.add_task(description="Upscaling", total=tiles_y * tiles_x)
+            with torch.no_grad():
+                for y in range(tiles_y):
+                    for x in range(tiles_x):
+                        # extract tile from input image
+                        ofs_x = x * self.tile_size
+                        ofs_y = y * self.tile_size
+                        # input tile area on total image
+                        input_start_x = ofs_x
+                        input_end_x = min(ofs_x + self.tile_size, width)
+                        input_start_y = ofs_y
+                        input_end_y = min(ofs_y + self.tile_size, height)
+
+                        # input tile area on total image with padding
+                        input_start_x_pad = max(input_start_x - self.tile_pad, 0)
+                        input_end_x_pad = min(input_end_x + self.tile_pad, width)
+                        input_start_y_pad = max(input_start_y - self.tile_pad, 0)
+                        input_end_y_pad = min(input_end_y + self.tile_pad, height)
+
+                        # input tile dimensions
+                        input_tile_width = input_end_x - input_start_x
+                        input_tile_height = input_end_y - input_start_y
+                        tile_idx = y * tiles_x + x + 1
+                        input_tile = self.img[:, :, input_start_y_pad:input_end_y_pad, input_start_x_pad:input_end_x_pad]
+
+                        # upscale tile
+                        try:
+                            output_tile = self.model(input_tile)
+                        except Exception as e:
+                            log.error(f'Upscale error: type=R-ESRGAN {e}')
+
+                        # output tile area on total image
+                        output_start_x = input_start_x * self.scale
+                        output_end_x = input_end_x * self.scale
+                        output_start_y = input_start_y * self.scale
+                        output_end_y = input_end_y * self.scale
+
+                        # output tile area without padding
+                        output_start_x_tile = (input_start_x - input_start_x_pad) * self.scale
+                        output_end_x_tile = output_start_x_tile + input_tile_width * self.scale
+                        output_start_y_tile = (input_start_y - input_start_y_pad) * self.scale
+                        output_end_y_tile = output_start_y_tile + input_tile_height * self.scale
+
+                        # put tile into output image
+                        self.output[:, :, output_start_y:output_end_y,
+                                    output_start_x:output_end_x] = output_tile[:, :, output_start_y_tile:output_end_y_tile,
+                                                                            output_start_x_tile:output_end_x_tile]
+                        progress.update(task, advance=1, description="Upscaling")
+
+    def post_process(self):
+        # remove extra pad
+        if self.mod_scale is not None:
+            _, _, h, w = self.output.size()
+            self.output = self.output[:, :, 0:h - self.mod_pad_h * self.scale, 0:w - self.mod_pad_w * self.scale]
+        # remove prepad
+        if self.pre_pad != 0:
+            _, _, h, w = self.output.size()
+            self.output = self.output[:, :, 0:h - self.pre_pad * self.scale, 0:w - self.pre_pad * self.scale]
+        return self.output
+
+    @torch.no_grad()
+    def enhance(self, img, outscale=None, alpha_upsampler='realesrgan'):
+        h_input, w_input = img.shape[0:2]
+        # img: numpy
+        img = img.astype(np.float32)
+        if np.max(img) > 256:  # 16-bit image
+            max_range = 65535
+        else:
+            max_range = 255
+        img = img / max_range
+        if len(img.shape) == 2:  # gray image
+            img_mode = 'L'
+            img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)
+        elif img.shape[2] == 4:  # RGBA image with alpha channel
+            img_mode = 'RGBA'
+            alpha = img[:, :, 3]
+            img = img[:, :, 0:3]
+            img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+            if alpha_upsampler == 'realesrgan':
+                alpha = cv2.cvtColor(alpha, cv2.COLOR_GRAY2RGB)
+        else:
+            img_mode = 'RGB'
+            img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+
+        # ------------------- process image (without the alpha channel) ------------------- #
+        self.pre_process(img)
+        if self.tile_size > 0:
+            self.tile_process()
+        else:
+            self.process()
+        output_img = self.post_process()
+        output_img = output_img.data.squeeze().float().cpu().clamp_(0, 1).numpy()
+        output_img = np.transpose(output_img[[2, 1, 0], :, :], (1, 2, 0))
+        if img_mode == 'L':
+            output_img = cv2.cvtColor(output_img, cv2.COLOR_BGR2GRAY)
+
+        # ------------------- process the alpha channel if necessary ------------------- #
+        if img_mode == 'RGBA':
+            if alpha_upsampler == 'realesrgan':
+                self.pre_process(alpha)
+                if self.tile_size > 0:
+                    self.tile_process()
+                else:
+                    self.process()
+                output_alpha = self.post_process()
+                output_alpha = output_alpha.data.squeeze().float().cpu().clamp_(0, 1).numpy()
+                output_alpha = np.transpose(output_alpha[[2, 1, 0], :, :], (1, 2, 0))
+                output_alpha = cv2.cvtColor(output_alpha, cv2.COLOR_BGR2GRAY)
+            else:  # use the cv2 resize for alpha channel
+                h, w = alpha.shape[0:2]
+                output_alpha = cv2.resize(alpha, (w * self.scale, h * self.scale), interpolation=cv2.INTER_LINEAR)
+
+            # merge the alpha channel
+            output_img = cv2.cvtColor(output_img, cv2.COLOR_BGR2BGRA)
+            output_img[:, :, 3] = output_alpha
+
+        # ------------------------------ return ------------------------------ #
+        if max_range == 65535:  # 16-bit image
+            output = (output_img * 65535.0).round().astype(np.uint16)
+        else:
+            output = (output_img * 255.0).round().astype(np.uint8)
+
+        if outscale is not None and outscale != float(self.scale):
+            output = cv2.resize(
+                output, (
+                    int(w_input * outscale),
+                    int(h_input * outscale),
+                ), interpolation=cv2.INTER_LANCZOS4)
+
+        return output, img_mode
+
+
+class PrefetchReader(threading.Thread):
+    """Prefetch images.
+
+    Args:
+        img_list (list[str]): A image list of image paths to be read.
+        num_prefetch_queue (int): Number of prefetch queue.
+    """
+
+    def __init__(self, img_list, num_prefetch_queue):
+        super().__init__()
+        self.que = queue.Queue(num_prefetch_queue)
+        self.img_list = img_list
+
+    def run(self):
+        for img_path in self.img_list:
+            img = cv2.imread(img_path, cv2.IMREAD_UNCHANGED)
+            self.que.put(img)
+
+        self.que.put(None)
+
+    def __next__(self):
+        next_item = self.que.get()
+        if next_item is None:
+            raise StopIteration
+        return next_item
+
+    def __iter__(self):
+        return self
+
+
+class IOConsumer(threading.Thread):
+
+    def __init__(self, opt, que, qid):
+        super().__init__()
+        self._queue = que
+        self.qid = qid
+        self.opt = opt
+
+    def run(self):
+        while True:
+            msg = self._queue.get()
+            if isinstance(msg, str) and msg == 'quit':
+                break
+
+            output = msg['output']
+            save_path = msg['save_path']
+            cv2.imwrite(save_path, output)
+
+from basicsr.utils.registry import ARCH_REGISTRY
+from torch import nn as nn
+from torch.nn import functional as F
+
+
+class SRVGGNetCompact(nn.Module):
+    """A compact VGG-style network structure for super-resolution.
+
+    It is a compact network structure, which performs upsampling in the last layer and no convolution is
+    conducted on the HR feature space.
+
+    Args:
+        num_in_ch (int): Channel number of inputs. Default: 3.
+        num_out_ch (int): Channel number of outputs. Default: 3.
+        num_feat (int): Channel number of intermediate features. Default: 64.
+        num_conv (int): Number of convolution layers in the body network. Default: 16.
+        upscale (int): Upsampling factor. Default: 4.
+        act_type (str): Activation type, options: 'relu', 'prelu', 'leakyrelu'. Default: prelu.
+    """
+
+    def __init__(self, num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=16, upscale=4, act_type='prelu'):
+        super(SRVGGNetCompact, self).__init__()
+        self.num_in_ch = num_in_ch
+        self.num_out_ch = num_out_ch
+        self.num_feat = num_feat
+        self.num_conv = num_conv
+        self.upscale = upscale
+        self.act_type = act_type
+
+        self.body = nn.ModuleList()
+        # the first conv
+        self.body.append(nn.Conv2d(num_in_ch, num_feat, 3, 1, 1))
+        # the first activation
+        if act_type == 'relu':
+            activation = nn.ReLU(inplace=True)
+        elif act_type == 'prelu':
+            activation = nn.PReLU(num_parameters=num_feat)
+        elif act_type == 'leakyrelu':
+            activation = nn.LeakyReLU(negative_slope=0.1, inplace=True)
+        self.body.append(activation)
+
+        # the body structure
+        for _ in range(num_conv):
+            self.body.append(nn.Conv2d(num_feat, num_feat, 3, 1, 1))
+            # activation
+            if act_type == 'relu':
+                activation = nn.ReLU(inplace=True)
+            elif act_type == 'prelu':
+                activation = nn.PReLU(num_parameters=num_feat)
+            elif act_type == 'leakyrelu':
+                activation = nn.LeakyReLU(negative_slope=0.1, inplace=True)
+            self.body.append(activation)
+
+        # the last conv
+        self.body.append(nn.Conv2d(num_feat, num_out_ch * upscale * upscale, 3, 1, 1))
+        # upsample
+        self.upsampler = nn.PixelShuffle(upscale)
+
+    def forward(self, x):
+        out = x
+        for i in range(0, len(self.body)):
+            out = self.body[i](out)
+
+        out = self.upsampler(out)
+        # add the nearest upsampled image, so that the network learns the residual
+        base = F.interpolate(x, scale_factor=self.upscale, mode='nearest')
+        out += base
+        return out
+    

modules/scripts.py

@@ -67,6 +67,20 @@ class Script:
         """
         pass # pylint: disable=unnecessary-pass
 
+    def setup(self, p, *args):
+        """For AlwaysVisible scripts, this function is called when the processing object is set up, before any processing starts.
+        args contains all values returned by components from ui().
+        """
+        pass # pylint: disable=unnecessary-pass
+
+    def before_process(self, p, *args):
+        """
+        This function is called very early during processing begins for AlwaysVisible scripts.
+        You can modify the processing object (p) here, inject hooks, etc.
+        args contains all values returned by components from ui()
+        """
+        pass # pylint: disable=unnecessary-pass
+
     def process(self, p, *args):
         """
         This function is called before processing begins for AlwaysVisible scripts.
@@ -398,7 +412,7 @@ class ScriptRunner:
 
         dropdown.init_field = init_field
         dropdown.change(fn=select_script, inputs=[dropdown], outputs=[script.group for script in self.selectable_scripts])
-        
+
         def onload_script_visibility(params):
             title = params.get('Script', None)
             if title:
@@ -437,6 +451,17 @@ class ScriptRunner:
         s.report()
         return processed
 
+    def before_process(self, p, **kwargs):
+        s = ScriptSummary('before-process')
+        for script in self.alwayson_scripts:
+            try:
+                script_args = p.script_args[script.args_from:script.args_to]
+                script.before_process(p, *script_args, **kwargs)
+            except Exception as e:
+                errors.display(e, f"Error running before process: {script.filename}")
+            s.record(script.title())
+        s.report()
+
     def process(self, p, **kwargs):
         s = ScriptSummary('process')
         for script in self.alwayson_scripts:

modules/scripts_postprocessing.py

@@ -1,6 +1,5 @@
 import os
 import gradio as gr
-
 from modules import errors, shared
 
 
@@ -15,15 +14,9 @@ class ScriptPostprocessing:
     controls = None
     args_from = None
     args_to = None
-
-    order = 1000
-    """scripts will be ordred by this value in postprocessing UI"""
-
-    name = None
-    """this function should return the title of the script."""
-
-    group = None
-    """A gr.Group component that has all script's UI inside it"""
+    order = 1000 # scripts will be ordred by this value in postprocessing UI
+    name = None # this function should return the title of the script
+    group = None # A gr.Group component that has all script's UI inside it
 
     def ui(self):
         """
@@ -61,25 +54,19 @@ class ScriptPostprocessingRunner:
 
     def initialize_scripts(self, scripts_data):
         self.scripts = []
-
         for script_class, path, _basedir, _script_module in scripts_data:
             script: ScriptPostprocessing = script_class()
             script.filename = path
-
             if script.name == "Simple Upscale":
                 continue
-
             self.scripts.append(script)
 
     def create_script_ui(self, script, inputs):
         script.args_from = len(inputs)
         script.args_to = len(inputs)
-
         script.controls = wrap_call(script.ui, script.filename, "ui")
-
         for control in script.controls.values():
             control.custom_script_source = os.path.basename(script.filename)
-
         inputs += list(script.controls.values())
         script.args_to = len(inputs)
 
@@ -87,56 +74,46 @@ class ScriptPostprocessingRunner:
         if self.scripts is None:
             import modules.scripts
             self.initialize_scripts(modules.scripts.postprocessing_scripts_data)
-
         scripts_order = shared.opts.postprocessing_operation_order
 
         def script_score(name):
             for i, possible_match in enumerate(scripts_order):
                 if possible_match == name:
                     return i
-
             return len(self.scripts)
 
         script_scores = {script.name: (script_score(script.name), script.order, script.name, original_index) for original_index, script in enumerate(self.scripts)}
-
         return sorted(self.scripts, key=lambda x: script_scores[x.name])
 
     def setup_ui(self):
         inputs = []
-
         for script in self.scripts_in_preferred_order():
             with gr.Row() as group:
                 self.create_script_ui(script, inputs)
-
             script.group = group
-
         self.ui_created = True
         return inputs
 
     def run(self, pp: PostprocessedImage, args):
         for script in self.scripts_in_preferred_order():
             shared.state.job = script.name
-
             script_args = args[script.args_from:script.args_to]
-
             process_args = {}
             for (name, _component), value in zip(script.controls.items(), script_args):
                 process_args[name] = value
-
+            shared.log.debug(f'postprocess: script={script.name} args={process_args}')
             script.process(pp, **process_args)
 
     def create_args_for_run(self, scripts_args):
         if not self.ui_created:
             with gr.Blocks(analytics_enabled=False):
                 self.setup_ui()
-
         scripts = self.scripts_in_preferred_order()
         args = [None] * max([x.args_to for x in scripts])
 
         for script in scripts:
             script_args_dict = scripts_args.get(script.name, None)
             if script_args_dict is not None:
-
                 for i, name in enumerate(script.controls):
                     args[script.args_from + i] = script_args_dict.get(name, None)
 

modules/sd_models.py

@@ -118,10 +118,13 @@ class CheckpointInfo:
 #Used by OpenVINO, can be used with TensorRT or Olive
 class CompiledModelState:
     def __init__(self):
+        self.first_pass = True
         self.height = 512
         self.width = 512
         self.batch_size = 1
-        self.first_pass = True
+        self.partition_id = 0
+        self.cn_model = []
+        self.lora_model = []
 
 
 class NoWatermark:
@@ -146,6 +149,7 @@ def checkpoint_tiles(use_short=False): # pylint: disable=unused-argument
 
 def list_models():
     t0 = time.time()
+    global checkpoints_list # pylint: disable=global-statement
     checkpoints_list.clear()
     checkpoint_aliases.clear()
     if shared.opts.sd_disable_ckpt or shared.backend == shared.Backend.DIFFUSERS:
@@ -172,6 +176,7 @@ def list_models():
         shared.log.warning(f"Checkpoint not found: {shared.cmd_opts.ckpt}")
     shared.log.info(f'Available models: {shared.opts.ckpt_dir} items={len(checkpoints_list)} time={time.time()-t0:.2f}s')
 
+    checkpoints_list = dict(sorted(checkpoints_list.items(), key=lambda cp: cp[1].filename))
     if len(checkpoints_list) == 0:
         if not shared.cmd_opts.no_download:
             key = input('Download the default model? (y/N) ')
@@ -190,6 +195,7 @@ def list_models():
                     if checkpoint_info.name is not None:
                         checkpoint_info.register()
 
+
 def update_model_hashes():
     txt = []
     lst = [ckpt for ckpt in checkpoints_list.values() if ckpt.hash is None]
@@ -586,8 +592,12 @@ model_data = ModelData()
 
 
 def change_backend():
-    shared.log.info(f'Pipeline changed: {shared.backend}')
-    unload_model_weights()
+    shared.log.info(f'Backend changed: {shared.backend}')
+    if shared.backend == shared.Backend.ORIGINAL:
+        change_from = shared.Backend.DIFFUSERS
+    else:
+        change_from = shared.Backend.ORIGINAL
+    unload_model_weights(change_from=change_from)
     checkpoints_loaded.clear()
     from modules.sd_samplers import list_samplers
     list_samplers(shared.backend)
@@ -654,6 +664,49 @@ def detect_pipeline(f: str, op: str = 'model'):
         pipeline = None, None
     return pipeline, guess
 
+def compile_diffusers(sd_model):
+    try:
+        if shared.opts.ipex_optimize:
+            import intel_extension_for_pytorch as ipex # pylint: disable=import-error, unused-import
+            sd_model.unet.training = False
+            sd_model.unet = ipex.optimize(sd_model.unet, dtype=devices.dtype_unet, inplace=True, weights_prepack=False) # pylint: disable=attribute-defined-outside-init
+            if hasattr(sd_model, 'vae'):
+                sd_model.vae.training = False
+                sd_model.vae = ipex.optimize(sd_model.vae, dtype=devices.dtype_vae, inplace=True, weights_prepack=False) # pylint: disable=attribute-defined-outside-init
+            if hasattr(sd_model, 'movq'):
+                sd_model.movq.training = False
+                sd_model.movq = ipex.optimize(sd_model.movq, dtype=devices.dtype_vae, inplace=True, weights_prepack=False) # pylint: disable=attribute-defined-outside-init
+            shared.log.info("Applied IPEX Optimize.")
+    except Exception as err:
+        shared.log.warning(f"IPEX Optimize not supported: {err}")
+
+    try:
+        if shared.opts.cuda_compile and shared.opts.cuda_compile_backend != 'none':
+            shared.log.info(f"Compiling pipeline={sd_model.__class__.__name__} shape={8 * sd_model.unet.config.sample_size} mode={shared.opts.cuda_compile_backend}")
+            import torch._dynamo # pylint: disable=unused-import,redefined-outer-name
+            if shared.opts.cuda_compile_backend == "openvino_fx":
+                torch._dynamo.reset() # pylint: disable=protected-access
+                from modules.intel.openvino import openvino_fx, openvino_clear_caches # pylint: disable=unused-import
+                openvino_clear_caches()
+                torch._dynamo.eval_frame.check_if_dynamo_supported = lambda: True # pylint: disable=protected-access
+                if shared.compiled_model_state is None:
+                    shared.compiled_model_state = CompiledModelState()
+                shared.compiled_model_state.first_pass = True if not shared.opts.cuda_compile_precompile else False
+            log_level = logging.WARNING if shared.opts.cuda_compile_verbose else logging.CRITICAL # pylint: disable=protected-access
+            if hasattr(torch, '_logging'):
+                torch._logging.set_logs(dynamo=log_level, aot=log_level, inductor=log_level) # pylint: disable=protected-access
+            torch._dynamo.config.verbose = shared.opts.cuda_compile_verbose # pylint: disable=protected-access
+            torch._dynamo.config.suppress_errors = shared.opts.cuda_compile_errors # pylint: disable=protected-access
+            sd_model.unet = torch.compile(sd_model.unet, mode=shared.opts.cuda_compile_mode, backend=shared.opts.cuda_compile_backend, fullgraph=shared.opts.cuda_compile_fullgraph) # pylint: disable=attribute-defined-outside-init
+            if hasattr(sd_model, 'vae'):
+                sd_model.vae.decode = torch.compile(sd_model.vae.decode, mode=shared.opts.cuda_compile_mode, backend=shared.opts.cuda_compile_backend, fullgraph=shared.opts.cuda_compile_fullgraph) # pylint: disable=attribute-defined-outside-init
+            if hasattr(sd_model, 'movq'):
+                sd_model.movq.decode = torch.compile(sd_model.movq.decode, mode=shared.opts.cuda_compile_mode, backend=shared.opts.cuda_compile_backend, fullgraph=shared.opts.cuda_compile_fullgraph) # pylint: disable=attribute-defined-outside-init
+            if shared.opts.cuda_compile_precompile:
+                sd_model("dummy prompt")
+            shared.log.info("Complilation done.")
+    except Exception as err:
+        shared.log.warning(f"Model compile not supported: {err}")
 
 def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=None, op='model'): # pylint: disable=unused-argument
     import torch # pylint: disable=reimported,redefined-outer-name
@@ -733,7 +786,7 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
                 shared.log.error(f'Failed loading {op}: {checkpoint_info.path} {e}')
                 return
         else:
-            diffusers_load_config["local_files_only "] = True
+            diffusers_load_config["local_files_only"] = True
             diffusers_load_config["extract_ema"] = shared.opts.diffusers_extract_ema
             pipeline, model_type = detect_pipeline(checkpoint_info.path, op)
             if pipeline is None:
@@ -762,7 +815,9 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
                     diffusers_load_config.pop('safety_checker', None)
                     diffusers_load_config.pop('requires_safety_checker', None)
                     diffusers_load_config.pop('load_safety_checker', None)
-                    shared.log.debug(f'Model {op}: pipeline={sd_model.__class__.__name__} config={diffusers_load_config}') # pylint: disable=protected-access
+                    diffusers_load_config.pop('config_files', None)
+                    diffusers_load_config.pop('local_files_only', None)
+                    shared.log.debug(f'Setting {op}: pipeline={sd_model.__class__.__name__} config={diffusers_load_config}') # pylint: disable=protected-access
             except Exception as e:
                 shared.log.error(f'Diffusers failed loading model using pipeline: {checkpoint_info.path} {shared.opts.diffusers_pipeline} {e}')
                 return
@@ -773,8 +828,8 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
             sd_model.scheduler.name = 'DDIM'
 
         if (shared.opts.diffusers_model_cpu_offload or shared.cmd_opts.medvram) and (shared.opts.diffusers_seq_cpu_offload or shared.cmd_opts.lowvram):
-            shared.log.warning(f'Model {op}: Model CPU offload (--medvram) and Sequential CPU offload (--lowvram) are not compatible')
-            shared.log.debug(f'Model {op}: disabling model CPU offload and --medvram')
+            shared.log.warning(f'Setting {op}: Model CPU offload and Sequential CPU offload are not compatible')
+            shared.log.debug(f'Setting {op}: disabling model CPU offload')
             shared.opts.diffusers_model_cpu_offload=False
             shared.cmd_opts.medvram=False
 
@@ -783,7 +838,7 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
         sd_model.has_accelerate = False
         if hasattr(sd_model, "enable_model_cpu_offload"):
             if (shared.cmd_opts.medvram and devices.backend != "directml") or shared.opts.diffusers_model_cpu_offload:
-                shared.log.debug(f'Model {op}: enable model CPU offload')
+                shared.log.debug(f'Setting {op}: enable model CPU offload')
                 if shared.opts.diffusers_move_base or shared.opts.diffusers_move_unet or shared.opts.diffusers_move_refiner:
                     shared.opts.diffusers_move_base = False
                     shared.opts.diffusers_move_unet = False
@@ -793,7 +848,7 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
                 sd_model.has_accelerate = True
         if hasattr(sd_model, "enable_sequential_cpu_offload"):
             if shared.cmd_opts.lowvram or shared.opts.diffusers_seq_cpu_offload:
-                shared.log.debug(f'Model {op}: enable sequential CPU offload')
+                shared.log.debug(f'Setting {op}: enable sequential CPU offload')
                 if shared.opts.diffusers_move_base or shared.opts.diffusers_move_unet or shared.opts.diffusers_move_refiner:
                     shared.opts.diffusers_move_base = False
                     shared.opts.diffusers_move_unet = False
@@ -803,19 +858,19 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
                 sd_model.has_accelerate = True
         if hasattr(sd_model, "enable_vae_slicing"):
             if shared.cmd_opts.lowvram or shared.opts.diffusers_vae_slicing:
-                shared.log.debug(f'Model {op}: enable VAE slicing')
+                shared.log.debug(f'Setting {op}: enable VAE slicing')
                 sd_model.enable_vae_slicing()
             else:
                 sd_model.disable_vae_slicing()
         if hasattr(sd_model, "enable_vae_tiling"):
             if shared.cmd_opts.lowvram or shared.opts.diffusers_vae_tiling:
-                shared.log.debug(f'Model {op}: enable VAE tiling')
+                shared.log.debug(f'Setting {op}: enable VAE tiling')
                 sd_model.enable_vae_tiling()
             else:
                 sd_model.disable_vae_tiling()
         if hasattr(sd_model, "enable_attention_slicing"):
             if shared.cmd_opts.lowvram or shared.opts.diffusers_attention_slicing:
-                shared.log.debug(f'Model {op}: enable attention slicing')
+                shared.log.debug(f'Setting {op}: enable attention slicing')
                 sd_model.enable_attention_slicing()
             else:
                 sd_model.disable_attention_slicing()
@@ -824,19 +879,19 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
                 sd_model.vae = vae
             if shared.opts.diffusers_vae_upcast != 'default':
                 if shared.opts.diffusers_vae_upcast == 'true':
-                    sd_model.vae.config["force_upcast"] = True
+                    # sd_model.vae.config["force_upcast"] = True
                     sd_model.vae.config.force_upcast = True
                 else:
-                    sd_model.vae.config["force_upcast"] = False
+                    # sd_model.vae.config["force_upcast"] = False
                     sd_model.vae.config.force_upcast = False
                 if shared.opts.no_half_vae:
                     devices.dtype_vae = torch.float32
                     sd_model.vae.to(devices.dtype_vae)
-            shared.log.debug(f'Model {op} VAE: name={sd_vae.loaded_vae_file} upcast={sd_model.vae.config.get("force_upcast", None)}')
+            shared.log.debug(f'Setting {op} VAE: name={sd_vae.loaded_vae_file} upcast={sd_model.vae.config.get("force_upcast", None)}')
         if shared.opts.cross_attention_optimization == "xFormers" and hasattr(sd_model, 'enable_xformers_memory_efficient_attention'):
             sd_model.enable_xformers_memory_efficient_attention()
         if shared.opts.opt_channelslast:
-            shared.log.debug(f'Model {op}: enable channels last')
+            shared.log.debug(f'Setting {op}: enable channels last')
             sd_model.unet.to(memory_format=torch.channels_last)
 
         base_sent_to_cpu=False
@@ -864,45 +919,8 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
                     base_sent_to_cpu=True
             elif not sd_model.has_accelerate:
                 sd_model.to(devices.device)
-            try:
-                if shared.opts.ipex_optimize:
-                    sd_model.unet.training = False
-                    sd_model.unet = torch.xpu.optimize(sd_model.unet, dtype=devices.dtype_unet, inplace=True, weights_prepack=False) # pylint: disable=attribute-defined-outside-init
-                    if hasattr(sd_model, 'vae'):
-                        sd_model.vae.training = False
-                        sd_model.vae = torch.xpu.optimize(sd_model.vae, dtype=devices.dtype_vae, inplace=True, weights_prepack=False) # pylint: disable=attribute-defined-outside-init
-                    if hasattr(sd_model, 'movq'):
-                        sd_model.movq.training = False
-                        sd_model.movq = torch.xpu.optimize(sd_model.movq, dtype=devices.dtype_vae, inplace=True, weights_prepack=False) # pylint: disable=attribute-defined-outside-init
-                    shared.log.info("Applied IPEX Optimize.")
-            except Exception as err:
-                shared.log.warning(f"IPEX Optimize not supported: {err}")
-            try:
-                if shared.opts.cuda_compile and shared.opts.cuda_compile_backend != 'none':
-                    shared.log.info(f"Compiling pipeline={sd_model.__class__.__name__} shape={8 * sd_model.unet.config.sample_size} mode={shared.opts.cuda_compile_backend}")
-                    import torch._dynamo # pylint: disable=unused-import,redefined-outer-name
-                    if shared.opts.cuda_compile_backend == "openvino_fx":
-                        torch._dynamo.reset() # pylint: disable=protected-access
-                        from modules.intel.openvino import openvino_fx, openvino_clear_caches # pylint: disable=unused-import
-                        openvino_clear_caches()
-                        torch._dynamo.eval_frame.check_if_dynamo_supported = lambda: True # pylint: disable=protected-access
-                        sd_model.compiled_model_state = CompiledModelState()
-                        sd_model.compiled_model_state.first_pass = True if not shared.opts.cuda_compile_precompile else False
-                    log_level = logging.WARNING if shared.opts.cuda_compile_verbose else logging.CRITICAL # pylint: disable=protected-access
-                    if hasattr(torch, '_logging'):
-                        torch._logging.set_logs(dynamo=log_level, aot=log_level, inductor=log_level) # pylint: disable=protected-access
-                    torch._dynamo.config.verbose = shared.opts.cuda_compile_verbose # pylint: disable=protected-access
-                    torch._dynamo.config.suppress_errors = shared.opts.cuda_compile_errors # pylint: disable=protected-access
-                    sd_model.unet = torch.compile(sd_model.unet, mode=shared.opts.cuda_compile_mode, backend=shared.opts.cuda_compile_backend, fullgraph=shared.opts.cuda_compile_fullgraph) # pylint: disable=attribute-defined-outside-init
-                    if hasattr(sd_model, 'vae'):
-                        sd_model.vae.decode = torch.compile(sd_model.vae.decode, mode=shared.opts.cuda_compile_mode, backend=shared.opts.cuda_compile_backend, fullgraph=shared.opts.cuda_compile_fullgraph) # pylint: disable=attribute-defined-outside-init
-                    if hasattr(sd_model, 'movq'):
-                        sd_model.movq.decode = torch.compile(sd_model.movq.decode, mode=shared.opts.cuda_compile_mode, backend=shared.opts.cuda_compile_backend, fullgraph=shared.opts.cuda_compile_fullgraph) # pylint: disable=attribute-defined-outside-init
-                    if shared.opts.cuda_compile_precompile:
-                        sd_model("dummy prompt")
-                    shared.log.info("Complilation done.")
-            except Exception as err:
-                shared.log.warning(f"Model compile not supported: {err}")
+
+            compile_diffusers(sd_model)
 
         if sd_model is None:
             shared.log.error('Diffuser model not loaded')
@@ -1162,11 +1180,11 @@ def disable_offload(sd_model):
         remove_hook_from_module(model, recurse=True)
 
 
-def unload_model_weights(op='model'):
-    from modules import sd_hijack
+def unload_model_weights(op='model', change_from='none'):
     if op == 'model' or op == 'dict':
         if model_data.sd_model:
-            if shared.backend == shared.Backend.ORIGINAL:
+            if (shared.backend == shared.Backend.ORIGINAL and change_from != shared.Backend.DIFFUSERS) or change_from == shared.Backend.ORIGINAL:
+                from modules import sd_hijack
                 model_data.sd_model.to(devices.cpu)
                 sd_hijack.model_hijack.undo_hijack(model_data.sd_model)
             else:
@@ -1176,7 +1194,8 @@ def unload_model_weights(op='model'):
             shared.log.debug(f'Unload weights {op}: {memory_stats()}')
     else:
         if model_data.sd_refiner:
-            if shared.backend == shared.Backend.ORIGINAL:
+            if (shared.backend == shared.Backend.ORIGINAL and change_from != shared.Backend.DIFFUSERS) or change_from == shared.Backend.ORIGINAL:
+                from modules import sd_hijack
                 model_data.sd_model.to(devices.cpu)
                 sd_hijack.model_hijack.undo_hijack(model_data.sd_refiner)
             else:
@@ -1208,5 +1227,7 @@ def apply_token_merging(sd_model, token_merging_ratio=0):
             )
             shared.log.debug(f'Applying token merging: ratio={token_merging_ratio}')
             sd_model.applied_token_merged_ratio = token_merging_ratio
-        except:
+        except Exception:
             shared.log.warning(f'Token merging not supported: pipeline={sd_model.__class__.__name__}')
+    else:
+        sd_model.applied_token_merged_ratio = 0

modules/sd_samplers_compvis.py

@@ -45,7 +45,6 @@ class VanillaStableDiffusionSampler:
     def launch_sampling(self, steps, func):
         state.sampling_steps = steps
         state.sampling_step = 0
-
         try:
             return func()
         except sd_samplers_common.InterruptedException:
@@ -53,11 +52,8 @@ class VanillaStableDiffusionSampler:
 
     def p_sample_ddim_hook(self, x_dec, cond, ts, unconditional_conditioning, *args, **kwargs):
         x_dec, ts, cond, unconditional_conditioning = self.before_sample(x_dec, ts, cond, unconditional_conditioning)
-
         res = self.orig_p_sample_ddim(x_dec, cond, ts, *args, unconditional_conditioning=unconditional_conditioning, **kwargs)
-
         x_dec, ts, cond, unconditional_conditioning, res = self.after_sample(x_dec, ts, cond, unconditional_conditioning, res)
-
         return res
 
     def before_sample(self, x, ts, cond, unconditional_conditioning):

modules/sd_samplers_diffusers.py

@@ -8,6 +8,7 @@ try:
         DEISMultistepScheduler,
         DPMSolverMultistepScheduler,
         DPMSolverSinglestepScheduler,
+        DPMSolverSDEScheduler,
         EulerAncestralDiscreteScheduler,
         EulerDiscreteScheduler,
         HeunDiscreteScheduler,
@@ -30,6 +31,7 @@ config = {
     'DEIS': { 'solver_order': 2, 'thresholding': False, 'sample_max_value': 1.0, 'algorithm_type': "deis", 'solver_type': "logrho", 'lower_order_final': True },
     'DPM 1S': { 'solver_order': 2, 'thresholding': False, 'sample_max_value': 1.0, 'algorithm_type': "dpmsolver++", 'solver_type': "midpoint", 'lower_order_final': True, 'use_karras_sigmas': False },
     'DPM 2M': { 'thresholding': False, 'sample_max_value': 1.0, 'algorithm_type': "dpmsolver++", 'solver_type': "midpoint", 'lower_order_final': True, 'use_karras_sigmas': False },
+    'DPM SDE': { 'use_karras_sigmas': False },
     'Euler a': { },
     'Euler': { 'interpolation_type': "linear", 'use_karras_sigmas': False },
     'Heun': { 'use_karras_sigmas': False },
@@ -52,6 +54,7 @@ samplers_data_diffusers = [
     sd_samplers_common.SamplerData('KDPM2 a', lambda model: DiffusionSampler('KDPM2 a', KDPM2AncestralDiscreteScheduler, model), [], {}),
     sd_samplers_common.SamplerData('DPM 1S', lambda model: DiffusionSampler('DPM++ 1S', DPMSolverSinglestepScheduler, model), [], {}),
     sd_samplers_common.SamplerData('DPM 2M', lambda model: DiffusionSampler('DPM++ 2M', DPMSolverMultistepScheduler, model), [], {}),
+    sd_samplers_common.SamplerData('DPM SDE', lambda model: DiffusionSampler('DPM SDE', DPMSolverSDEScheduler, model), [], {}),
     sd_samplers_common.SamplerData('Euler', lambda model: DiffusionSampler('Euler', EulerDiscreteScheduler, model), [], {}),
     sd_samplers_common.SamplerData('Euler a', lambda model: DiffusionSampler('Euler a', EulerAncestralDiscreteScheduler, model), [], {}),
     sd_samplers_common.SamplerData('Heun', lambda model: DiffusionSampler('Heun', HeunDiscreteScheduler, model), [], {}),

modules/shared.py

@@ -9,10 +9,9 @@ import urllib.request
 from urllib.parse import urlparse
 from enum import Enum
 import gradio as gr
-import tqdm
 import fasteners
 from rich.console import Console
-from modules import errors, ui_components, shared_items, cmd_args
+from modules import errors, shared_items, cmd_args, ui_components
 from modules.paths_internal import models_path, script_path, data_path, sd_configs_path, sd_default_config, sd_model_file, default_sd_model_file, extensions_dir, extensions_builtin_dir # pylint: disable=W0611
 from modules.dml import memory_providers, default_memory_provider, directml_do_hijack
 import modules.interrogate
@@ -38,6 +37,7 @@ interrogator = modules.interrogate.InterrogateModels("interrogate")
 sd_upscalers = []
 face_restorers = []
 tab_names = []
+extra_networks = []
 options_templates = {}
 hypernetworks = {}
 loaded_hypernetworks = []
@@ -72,21 +72,12 @@ restricted_opts = {
 compatibility_opts = ['clip_skip', 'uni_pc_lower_order_final', 'uni_pc_order']
 console = Console(log_time=True, log_time_format='%H:%M:%S-%f')
 
-def is_url(string):
-    parsed_url = urlparse(string)
-    return all([parsed_url.scheme, parsed_url.netloc])
-
 
 class Backend(Enum):
     ORIGINAL = 1
     DIFFUSERS = 2
 
 
-def reload_hypernetworks():
-    from modules.hypernetworks import hypernetwork
-    global hypernetworks # pylint: disable=W0603
-    hypernetworks = hypernetwork.list_hypernetworks(opts.hypernetwork_dir)
-
 
 class State:
     skipped = False
@@ -174,7 +165,7 @@ class State:
         """sets self.current_image from self.current_latent if enough sampling steps have been made after the last call to this"""
         if not parallel_processing_allowed:
             return
-        if self.sampling_step - self.current_image_sampling_step >= opts.show_progress_every_n_steps and opts.live_previews_enable and opts.show_progress_every_n_steps != -1:
+        if abs(self.sampling_step - self.current_image_sampling_step) >= opts.show_progress_every_n_steps and opts.live_previews_enable and opts.show_progress_every_n_steps > 0:
             self.do_set_current_image()
 
     def do_set_current_image(self):
@@ -193,6 +184,7 @@ class State:
         self.current_image = image
         self.id_live_preview += 1
 
+
 state = State()
 state.server_start = time.time()
 if not hasattr(cmd_opts, "use_openvino"):
@@ -252,6 +244,17 @@ def list_checkpoint_tiles():
 default_checkpoint = list_checkpoint_tiles()[0] if len(list_checkpoint_tiles()) > 0 else "model.ckpt"
 
 
+def is_url(string):
+    parsed_url = urlparse(string)
+    return all([parsed_url.scheme, parsed_url.netloc])
+
+
+def reload_hypernetworks():
+    from modules.hypernetworks import hypernetwork
+    global hypernetworks # pylint: disable=W0603
+    hypernetworks = hypernetwork.list_hypernetworks(opts.hypernetwork_dir)
+
+
 def refresh_checkpoints():
     import modules.sd_models # pylint: disable=W0621
     return modules.sd_models.list_models()
@@ -435,10 +438,10 @@ options_templates.update(options_section(('diffusers', "Diffusers Settings"), {
     "diffusers_vae_upcast": OptionInfo("default", "VAE upcasting", gr.Radio, lambda: {"choices": ['default', 'true', 'false']}),
     "diffusers_vae_slicing": OptionInfo(True, "Enable VAE slicing"),
     "diffusers_vae_tiling": OptionInfo(False if cmd_opts.use_openvino else True, "Enable VAE tiling"),
-    "diffusers_attention_slicing": OptionInfo(False, "Enable attention slicing"),
+    "diffusers_attention_slicing": OptionInfo(True if devices.backend == "ipex" else False, "Enable attention slicing"),
     "diffusers_model_load_variant": OptionInfo("default", "Diffusers model loading variant", gr.Radio, lambda: {"choices": ['default', 'fp32', 'fp16']}),
     "diffusers_vae_load_variant": OptionInfo("default", "Diffusers VAE loading variant", gr.Radio, lambda: {"choices": ['default', 'fp32', 'fp16']}),
-    "diffusers_lora_loader": OptionInfo("diffusers default" if cmd_opts.use_openvino else "sequential apply", "Diffusers LoRA loading variant", gr.Radio, lambda: {"choices": ['sequential apply', 'merge and apply', 'diffusers default']}),
+    "diffusers_lora_loader": OptionInfo("sequential apply", "Diffusers LoRA loading variant", gr.Radio, lambda: {"choices": ['diffusers', 'sequential apply', 'merge and apply']}),
     "diffusers_force_zeros": OptionInfo(True, "Force zeros for prompts when empty"),
     "diffusers_aesthetics_score": OptionInfo(False, "Require aesthetics score"),
 }))
@@ -487,14 +490,13 @@ options_templates.update(options_section(('saving-images', "Image Options"), {
     "n_rows": OptionInfo(-1, "Grid row count", gr.Slider, {"minimum": -1, "maximum": 16, "step": 1}),
 
     "save_sep_options": OptionInfo("<h2>Intermediate Image Saving</h2>", "", gr.HTML),
-    "save_init_img": OptionInfo(True, "Save copy of img2img init images"),
-    "save_images_before_highres_fix": OptionInfo(False, "Save copy of image before applying highres fix"),
+    "save_init_img": OptionInfo(False, "Save copy of img2img init images"),
+    "save_images_before_highres_fix": OptionInfo(False, "Save copy of image before applying hires"),
     "save_images_before_refiner": OptionInfo(False, "Save copy of image before running refiner"),
     "save_images_before_face_restoration": OptionInfo(False, "Save copy of image before doing face restoration"),
     "save_images_before_color_correction": OptionInfo(False, "Save copy of image before applying color correction"),
     "save_mask": OptionInfo(False, "Save copy of the inpainting greyscale mask"),
     "save_mask_composite": OptionInfo(False, "Save copy of inpainting masked composite"),
-
 }))
 
 options_templates.update(options_section(('saving-paths', "Image Naming & Paths"), {
@@ -522,7 +524,6 @@ options_templates.update(options_section(('saving-paths', "Image Naming & Paths"
     "outdir_grids": OptionInfo("", "Output directory for grids", component_args=hide_dirs, folder=True),
     "outdir_txt2img_grids": OptionInfo("outputs/grids", 'Output directory for txt2img grids', component_args=hide_dirs, folder=True),
     "outdir_img2img_grids": OptionInfo("outputs/grids", 'Output directory for img2img grids', component_args=hide_dirs, folder=True),
-
 }))
 
 options_templates.update(options_section(('ui', "User Interface"), {
@@ -558,7 +559,7 @@ options_templates.update(options_section(('live-preview', "Live Previews"), {
 }))
 
 options_templates.update(options_section(('sampler-params', "Sampler Settings"), {
-    "show_samplers": OptionInfo(["Default", "Euler a", "UniPC", "DEIS", "DDIM", "DPM 1S", "DPM 2M", "DPM++ 2M SDE", "DPM++ 2M SDE Karras", "DPM2 Karras", "DPM++ 2M Karras"], "Show samplers in user interface", gr.CheckboxGroup, lambda: {"choices": [x.name for x in list_samplers() if x.name != "PLMS"]}),
+    "show_samplers": OptionInfo(["Default", "Euler a", "UniPC", "DEIS", "DDIM", "DPM 1S", "DPM 2M", "DPM SDE", "DPM++ 2M SDE", "DPM++ 2M SDE Karras", "DPM2 Karras", "DPM++ 2M Karras"], "Show samplers in user interface", gr.CheckboxGroup, lambda: {"choices": [x.name for x in list_samplers() if x.name != "PLMS"]}),
     'uni_pc_variant': OptionInfo("bh1", "UniPC variant", gr.Radio, {"choices": ["bh1", "bh2", "vary_coeff"]}),
     'uni_pc_skip_type': OptionInfo("time_uniform", "UniPC skip type", gr.Radio, {"choices": ["time_uniform", "time_quadratic", "logSNR"]}),
     'eta_noise_seed_delta': OptionInfo(0, "Noise seed delta (eta)", gr.Number, {"precision": 0}),
@@ -570,7 +571,7 @@ options_templates.update(options_section(('sampler-params', "Sampler Settings"),
     "schedulers_use_karras": OptionInfo(True, "Samplers use Karras sigmas where applicable"),
     "schedulers_use_loworder": OptionInfo(True, "Samplers use simplified solvers in final steps where applicable"),
     "schedulers_use_thresholding": OptionInfo(False, "Samplers use dynamic thresholding where applicable"),
-    "schedulers_dpm_solver": OptionInfo("sde-dpmsolver++", "Samplers DPM solver algorithm", gr.Radio, lambda: {"choices": ['dpmsolver', 'dpmsolver++', 'sde-dpmsolver++']}),
+    "schedulers_dpm_solver": OptionInfo("sde-dpmsolver++", "Samplers DPM solver algorithm", gr.Radio, lambda: {"choices": ['dpmsolver', 'dpmsolver++', 'sde-dpmsolver', 'sde-dpmsolver++']}),
     "schedulers_beta_schedule": OptionInfo("default", "Samplers override beta schedule", gr.Radio, lambda: {"choices": ['default', 'linear', 'scaled_linear', 'squaredcos_cap_v2']}),
     'schedulers_beta_start': OptionInfo(0, "Samplers override beta start", gr.Number, {}),
     'schedulers_beta_end': OptionInfo(0, "Samplers override beta end", gr.Number, {}),
@@ -609,7 +610,7 @@ options_templates.update(options_section(('postprocessing', "Postprocessing"), {
     "postprocessing_sep_upscalers": OptionInfo("<h2>Upscaling</h2>", "", gr.HTML),
     'upscaling_max_images_in_cache': OptionInfo(5, "Maximum number of images in upscaling cache", gr.Slider, {"minimum": 0, "maximum": 10, "step": 1}),
     "upscaler_for_img2img": OptionInfo("None", "Default upscaler for image resize operations", gr.Dropdown, lambda: {"choices": [x.name for x in sd_upscalers]}),
-    "realesrgan_enabled_models": OptionInfo(["R-ESRGAN 4x+", "R-ESRGAN 4x+ Anime6B"], "Real-ESRGAN available models", gr.CheckboxGroup, lambda: {"choices": shared_items.realesrgan_models_names()}),
+    # "realesrgan_enabled_models": OptionInfo(["R-ESRGAN 4x+", "R-ESRGAN 4x+ Anime6B"], "Real-ESRGAN available models", gr.CheckboxGroup, lambda: {"choices": shared_items.realesrgan_models_names()}),
     "ESRGAN_tile": OptionInfo(192, "Tile size for ESRGAN upscalers", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}),
     "ESRGAN_tile_overlap": OptionInfo(8, "Tile overlap in pixels for ESRGAN upscalers", gr.Slider, {"minimum": 0, "maximum": 48, "step": 1}),
     "SCUNET_tile": OptionInfo(256, "Tile size for SCUNET upscalers", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}),
@@ -647,6 +648,7 @@ options_templates.update(options_section(('interrogate', "Interrogate"), {
 }))
 
 options_templates.update(options_section(('extra_networks', "Extra Networks"), {
+    "extra_networks": OptionInfo(["All"], "Extra networks", ui_components.DropdownMulti, lambda: {"choices": ['All'] + [en.title for en in extra_networks]}),
     "extra_networks_card_cover": OptionInfo("sidebar", "UI position", gr.Radio, lambda: {"choices": ["cover", "inline", "sidebar"]}),
     "extra_networks_height": OptionInfo(53, "UI height (%)", gr.Slider, {"minimum": 10, "maximum": 100, "step": 1}),
     "extra_networks_sidebar_width": OptionInfo(35, "UI sidebar width (%)", gr.Slider, {"minimum": 10, "maximum": 80, "step": 1}),
@@ -656,8 +658,8 @@ options_templates.update(options_section(('extra_networks', "Extra Networks"), {
     "extra_networks_card_fit": OptionInfo("cover", "UI image contain method", gr.Radio, lambda: {"choices": ["contain", "cover", "fill"]}),
     "extra_network_skip_indexing": OptionInfo(False, "Do not automatically build extra network pages", gr.Checkbox),
     "lyco_patch_lora": OptionInfo(False, "Use LyCoris handler for all LoRA types", gr.Checkbox),
-    "lora_functional": OptionInfo(False, "Use Kohya method for handling multiple LoRA", gr.Checkbox),
-    "extra_networks_default_multiplier": OptionInfo(1.0, "Multiplier for extra networks", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01}),
+    "lora_functional": OptionInfo(False, "Use Kohya method for handling multiple LoRA", gr.Checkbox, { "visible": False }),
+    "extra_networks_default_multiplier": OptionInfo(1.0, "Multiplier for extra networks", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01, "visible": False}),
     "sd_hypernetwork": OptionInfo("None", "Add hypernetwork to prompt", gr.Dropdown, lambda: { "choices": ["None"] + list(hypernetworks.keys()), "visible": False }, refresh=reload_hypernetworks),
 }))
 
@@ -832,6 +834,7 @@ else:
 opts.data['sd_backend'] = 'diffusers' if backend == Backend.DIFFUSERS else 'original'
 opts.data['uni_pc_lower_order_final'] = opts.schedulers_use_loworder
 opts.data['uni_pc_order'] = opts.schedulers_solver_order
+# opts.data['diffusers_lora_loader'] = 'diffusers' # TODO broken in diffusers=0.21
 log.info(f'Engine: backend={backend} compute={devices.backend} mode={devices.inference_context.__name__} device={devices.get_optimal_device_name()}')
 log.info(f'Device: {print_dict(devices.get_gpu_info())}')
 
@@ -885,37 +888,18 @@ def reload_gradio_theme(theme_name=None):
     log.info(f'Loading UI theme: name={theme_name} style={opts.theme_style}')
 
 
-class TotalTQDM:
+class TotalTQDM: # compatibility with previous global-tqdm
+    # import tqdm
     def __init__(self):
-        self._tqdm = None
-
+        pass
     def reset(self):
-        self._tqdm = tqdm.tqdm(
-            desc="Total",
-            total=state.job_count * state.sampling_steps,
-            position=1,
-        )
-
+        pass
     def update(self):
-        if not opts.multiple_tqdm or cmd_opts.disable_console_progressbars:
-            return
-        if self._tqdm is None:
-            self.reset()
-        self._tqdm.update()
-
+        pass
     def updateTotal(self, new_total):
-        if not opts.multiple_tqdm or cmd_opts.disable_console_progressbars:
-            return
-        if self._tqdm is None:
-            self.reset()
-        self._tqdm.total = new_total
-
+        pass
     def clear(self):
-        if self._tqdm is not None:
-            self._tqdm.refresh()
-            self._tqdm.close()
-            self._tqdm = None
-
+        pass
 total_tqdm = TotalTQDM()
 
 
@@ -1067,4 +1051,5 @@ sd_model = None
 sd_refiner = None
 sd_model_type = ''
 sd_refiner_type = ''
+compiled_model_state = None
 sys.modules[__name__].__class__ = Shared

modules/styles.py

@@ -1,7 +1,8 @@
 # We need this so Python doesn't complain about the unknown StableDiffusionProcessing-typehint at runtime
 from __future__ import annotations
-import csv
+import re
 import os
+import csv
 import json
 from installer import log
 from modules import paths
@@ -9,7 +10,7 @@ from modules import paths
 
 class Style():
     def __init__(self, name: str, prompt: str = "", negative_prompt: str = "", extra: str = "", filename: str = "", preview: str = ""):
-        self.name = name
+        self.name = re.sub(r'[\t\r\n]', '', name).strip()
         self.prompt = prompt
         self.negative_prompt = negative_prompt
         self.extra = extra
@@ -73,6 +74,7 @@ class StyleDatabase:
                     list_folder(fn)
 
         list_folder(self.path)
+        self.styles = dict(sorted(self.styles.items(), key=lambda style: style[1].filename))
         log.debug(f'Loaded styles: folder={self.path} items={len(self.styles.keys())}')
 
     def get_style_prompts(self, styles):

modules/textual_inversion/textual_inversion.py

@@ -159,11 +159,6 @@ class EmbeddingDatabase:
             self.register_embedding(embedding, shared.sd_model)
         except Exception:
             self.skipped_embeddings[name] = embedding
-        try:
-            text_inv_tokens = pipe.tokenizer.added_tokens_encoder.keys()
-            text_inv_tokens = [t for t in text_inv_tokens if not (len(t.split("_")) > 1 and t.split("_")[-1].isdigit())]
-        except Exception:
-            text_inv_tokens = []
 
     def load_from_file(self, path, filename):
         name, ext = os.path.splitext(filename)

modules/ui.py

@@ -937,7 +937,7 @@ def create_ui(startup_timer = None):
         elif info.folder is not None:
             with FormRow():
                 res = comp(label=info.label, value=fun(), elem_id=elem_id, elem_classes="folder-selector", **args)
-                ui_common.create_browse_button(res, f"folder_{key}")
+                # ui_common.create_browse_button(res, f"folder_{key}")
         else:
             try:
                 res = comp(label=info.label, value=fun(), elem_id=elem_id, **args)
@@ -1076,6 +1076,11 @@ def create_ui(startup_timer = None):
                 loadsave.create_ui()
                 create_dirty_indicator("tab_defaults", [], interactive=False)
 
+            with gr.TabItem("Change log", id="change_log", elem_id="system_tab_changelog"):
+                with open('CHANGELOG.md', 'r', encoding='utf-8') as f:
+                    md = f.read()
+                gr.Markdown(md)
+
             with gr.TabItem("Licenses", id="system_licenses", elem_id="system_tab_licenses"):
                 gr.HTML(modules.shared.html("licenses.html"), elem_id="licenses", elem_classes="licenses")
                 create_dirty_indicator("tab_licenses", [], interactive=False)

modules/ui_common.py

@@ -110,7 +110,7 @@ def save_files(js_data, images, html_info, index):
             fullfns.append(fullfn)
             destination = shared.opts.outdir_save
             if shared.opts.use_save_to_dirs_for_ui:
-                namegen = modules.images.FilenameGenerator(p, seed=p.all_seeds[i], prompt=p.all_prompts[i], image=None)  # pylint: disable=no-member
+                namegen = modules.images.FilenameGenerator(p, seed=p.all_seeds[i], prompt=p.all_prompts[i], image=None, index=image_index)  # pylint: disable=no-member
                 dirname = namegen.apply(shared.opts.directories_filename_pattern or "[prompt_words]").lstrip(' ').rstrip('\\ /')
                 destination = os.path.join(destination, dirname)
                 os.makedirs(destination, exist_ok = True)

modules/ui_extra_networks.py

@@ -16,10 +16,10 @@ from modules.ui_components import ToolButton
 import modules.ui_symbols as symbols
 
 
-extra_pages = []
 allowed_dirs = []
 dir_cache = {} # key=path, value=(mtime, listdir(path))
 refresh_time = None
+extra_pages = shared.extra_networks
 
 
 def listdir(path):
@@ -37,9 +37,9 @@ def listdir(path):
 
 def register_page(page):
     # registers extra networks page for the UI; recommend doing it in on_before_ui() callback for extensions
-    extra_pages.append(page)
+    shared.extra_networks.append(page)
     allowed_dirs.clear()
-    for page in extra_pages:
+    for page in shared.extra_networks:
         for folder in page.allowed_directories_for_previews():
             if folder not in allowed_dirs:
                 allowed_dirs.append(os.path.abspath(folder))
@@ -58,7 +58,7 @@ def fetch_file(filename: str = ""):
 
 
 def get_metadata(page: str = "", item: str = ""):
-    page = next(iter([x for x in extra_pages if x.name == page]), None)
+    page = next(iter([x for x in shared.extra_networks if x.name == page]), None)
     if page is None:
         return JSONResponse({ 'metadata': 'none' })
     metadata = page.metadata.get(item, 'none')
@@ -69,7 +69,7 @@ def get_metadata(page: str = "", item: str = ""):
 
 
 def get_info(page: str = "", item: str = ""):
-    page = next(iter([x for x in extra_pages if x.name == page]), None)
+    page = next(iter([x for x in shared.extra_networks if x.name == page]), None)
     if page is None:
         return JSONResponse({ 'info': 'none' })
     info = page.info.get(item, 'none')
@@ -128,7 +128,7 @@ class ExtraNetworksPage:
                     shared.log.error(f'Cannot evaluate extra network prompt: {item["prompt"]} {e}')
 
         if not any(self.title in x.label for x in xyz_grid.axis_options):
-            if self.title == 'Checkpoints':
+            if self.title == 'Model':
                 return
             opt = xyz_grid.AxisOption(f"[Network] {self.title}", str, add_prompt, choices=lambda: [x["name"] for x in self.items])
             xyz_grid.axis_options.append(opt)
@@ -308,7 +308,7 @@ class ExtraNetworksPage:
             shared.log.error(f'Extra network save preview: {filename} {e}')
             return
         is_allowed = False
-        for page in extra_pages:
+        for page in shared.extra_networks:
             if any(path_is_parent(x, filename) for x in page.allowed_directories_for_previews()):
                 is_allowed = True
                 break
@@ -339,7 +339,7 @@ class ExtraNetworksPage:
 
 
 def initialize():
-    extra_pages.clear()
+    shared.extra_networks.clear()
 
 
 def register_pages():
@@ -353,6 +353,21 @@ def register_pages():
     register_page(ExtraNetworksPageHypernetworks())
 
 
+def get_pages():
+    pages = []
+    if 'All' in shared.opts.extra_networks:
+        pages = shared.extra_networks
+    else:
+        titles = [page.title for page in shared.extra_networks]
+        for page in shared.opts.extra_networks:
+            try:
+                idx = titles.index(page)
+            except ValueError:
+                continue
+            pages.append(shared.extra_networks[idx])
+    return pages
+
+
 class ExtraNetworksUi:
     def __init__(self):
         self.pages = None
@@ -384,28 +399,28 @@ def create_ui(container, button, tabname, skip_indexing = False):
         if ui.tabname == 'txt2img': # refresh only once
             global refresh_time # pylint: disable=global-statement
             refresh_time = time.time()
-        for page in extra_pages:
+        for page in get_pages():
             page.create_page(ui.tabname, skip_indexing)
             with gr.Tab(page.title, id=page.title.lower().replace(" ", "_"), elem_classes="extra-networks-tab"):
-                page_elem = gr.HTML(page.html, elem_id=f'{tabname}{page.name}_extra_page', elem_classes="extra-networks-page")
-                page_elem.change(fn=lambda: None, _js=f'() => refreshExtraNetworks("{tabname}")', inputs=[], outputs=[])
-                ui.pages.append(page_elem)
+                hmtl = gr.HTML(page.html, elem_id=f'{tabname}{page.name}_extra_page', elem_classes="extra-networks-page")
+                # hmtl.change(fn=lambda: None, _js=f'() => refreshExtraNetworks("{tabname}")', inputs=[], outputs=[])
+                ui.pages.append(hmtl)
 
     def toggle_visibility(is_visible):
         is_visible = not is_visible
         return is_visible, gr.update(visible=is_visible), gr.update(variant=("secondary-down" if is_visible else "secondary"))
 
     def en_refresh(title):
-        res = []
-        for page in extra_pages:
+        pages = []
+        for page in get_pages():
             if title is None or title == '' or title == page.title or len(page.html) == 0:
                 page.refresh()
                 page.refresh_time = None
                 page.create_page(ui.tabname)
                 shared.log.debug(f"Refreshing Extra networks: page='{page.title}' items={len(page.items)} tab={ui.tabname}")
-            res.append(page.html)
+            pages.append(page.html)
         ui.search.update(value = ui.search.value)
-        return res
+        return pages
 
     state_visible = gr.State(value=False) # pylint: disable=abstract-class-instantiated
     button.click(fn=toggle_visibility, inputs=[state_visible], outputs=[state_visible, container, button])
@@ -423,15 +438,13 @@ def path_is_parent(parent_path, child_path):
 def setup_ui(ui, gallery):
 
     def save_preview(pagename, index, images, filename):
-        res = []
-        for page in extra_pages:
+        pages = []
+        for page in get_pages():
             if pagename is None or pagename == '' or pagename == page.title or len(page.html) == 0:
                 page.save_preview(index, images, filename)
-                res.append(page.create_page(ui.tabname))
-            else:
-                res.append(page.html)
-        return res
-
+                page.create_page(ui.tabname)
+            pages.append(page.html)
+        return pages
 
     ui.button_save_preview.click(
         fn=save_preview,
@@ -441,14 +454,13 @@ def setup_ui(ui, gallery):
     )
 
     def save_description(pagename, filename, desc):
-        res = []
-        for page in extra_pages:
+        pages = []
+        for page in get_pages():
             if pagename is None or pagename == '' or pagename == page.title or len(page.html) == 0:
                 page.save_description(filename, desc)
-                res.append(page.create_page(ui.tabname))
-            else:
-                res.append(page.html)
-        return res
+                page.create_page(ui.tabname)
+            pages.append(page.html)
+        return pages
 
     ui.button_save_description.click(
         fn=save_description,

modules/ui_extra_networks_checkpoints.py

@@ -7,7 +7,7 @@ from modules import shared, ui_extra_networks, sd_models
 
 class ExtraNetworksPageCheckpoints(ui_extra_networks.ExtraNetworksPage):
     def __init__(self):
-        super().__init__('Checkpoints')
+        super().__init__('Model')
 
     def refresh(self):
         shared.refresh_checkpoints()
@@ -18,6 +18,7 @@ class ExtraNetworksPageCheckpoints(ui_extra_networks.ExtraNetworksPage):
             path, _ext = os.path.splitext(checkpoint.filename)
             yield {
                 "name": checkpoint.name_for_extra,
+                "title": checkpoint.title,
                 "filename": path,
                 "fullname": checkpoint.filename,
                 "hash": checkpoint.shorthash,

modules/ui_extra_networks_hypernets.py

@@ -5,7 +5,7 @@ from modules import shared, ui_extra_networks
 
 class ExtraNetworksPageHypernetworks(ui_extra_networks.ExtraNetworksPage):
     def __init__(self):
-        super().__init__('Hypernetworks')
+        super().__init__('Hypernetwork')
 
     def refresh(self):
         shared.reload_hypernetworks()

modules/ui_extra_networks_styles.py

@@ -6,7 +6,7 @@ from modules import shared, ui_extra_networks
 
 class ExtraNetworksPageStyles(ui_extra_networks.ExtraNetworksPage):
     def __init__(self):
-        super().__init__('Styles')
+        super().__init__('Style')
 
     def refresh(self):
         shared.prompt_styles.reload()

modules/ui_extra_networks_textual_inversion.py

@@ -7,7 +7,7 @@ from modules.textual_inversion.textual_inversion import Embedding
 
 class ExtraNetworksPageTextualInversion(ui_extra_networks.ExtraNetworksPage):
     def __init__(self):
-        super().__init__('Textual Inversion')
+        super().__init__('Embedding')
         self.allow_negative_prompt = True
 
     def refresh(self):
@@ -39,6 +39,7 @@ class ExtraNetworksPageTextualInversion(ui_extra_networks.ExtraNetworksPage):
             embeddings = list(sd_models.model_data.sd_model.embedding_db.word_embeddings.values())
         else:
             embeddings = []
+        embeddings = sorted(embeddings, key=lambda emb: emb.filename)
         for embedding in embeddings:
             path, _ext = os.path.splitext(embedding.filename)
             tags = {}

requirements.txt

@@ -27,7 +27,6 @@ opencv-contrib-python-headless
 piexif
 psutil
 pyyaml
-realesrgan
 resize-right
 rich
 safetensors
@@ -47,7 +46,7 @@ requests==2.31.0
 tqdm==4.66.1
 accelerate==0.20.3
 opencv-python-headless==4.7.0.72
-diffusers==0.21.1
+diffusers==0.20.2
 einops==0.4.1
 gradio==3.43.2
 huggingface_hub==0.17.1

webui.bat

@@ -28,7 +28,7 @@ if %ERRORLEVEL% == 0 goto :activate_venv
 
 for /f "delims=" %%i in ('CALL %PYTHON% -c "import sys; print(sys.executable)"') do set PYTHON_FULLNAME="%%i"
 echo Using python: %PYTHON_FULLNAME%
-echo Creating VENV: %VENV_DIR% 
+echo Creating VENV: %VENV_DIR%
 %PYTHON_FULLNAME% -m venv "%VENV_DIR%" >tmp/stdout.txt 2>tmp/stderr.txt
 if %ERRORLEVEL% == 0 goto :activate_venv
 echo Failed creating VENV: "%VENV_DIR%"