Merge pull request #2401 from vladmandic/master

update dev

CHANGELOG.md

@@ -5,33 +5,31 @@
 Service release addressing all zero-day issues reported so far...
 
 **Fixes**
-- fix freeu for backend original and add it to xyz grid
+- fix **freeu** for backend original and add it to xyz grid
 - fix loading diffuser models in huggingface format from non-standard location
 - fix default styles looking in wrong location
 - fix missing upscaler folder on initial startup
 - fix handling of relative path for models
 - fix simple live preview device mismatch
 - fix batch img2img
-- fix diffusers dpm++ 2m, dpm++ 1s, deis samplers
+- fix diffusers samplers: dpm++ 2m, dpm++ 1s, deis
 - fix new style filename template
 - fix image name template using model name
 - fix model path using relative path
 - fix `torch-rocm` and `tensorflow-rocm` version detection, thanks @xangelix
-- fix chainner upscalers color clipping
+- fix **chainner** upscalers color clipping
 - fix for base+refiner workflow in diffusers mode: number of steps, diffuser pipe mode  
 - fix for prompt encoder with refiner in diffusers mode
 - fix prompts-from-file saving incorrect metadata
 - fix before-hires step
 - fix diffusers switch from invalid model
-- directml and ipex updates
+- **directml** and **ipex** updates
 - force second requirements check on startup
 - remove lyco, multiple_tqdm
 - enhance extension compatibility for exensions directly importing codeformers  
 - enhance extension compatibility for exensions directly accessing processing params  
 - css fixes
 - clearly mark external themes in ui
-- new option: *settings -> images -> keep incomplete*
-  can be used to skip vae decode on aborted/skipped/interrupted image generations
 - update `openvino`, thanks @disty0
 - update `typing-extensions`
 
@@ -39,7 +37,12 @@ Service release addressing all zero-day issues reported so far...
 
 - remove external clone of items in `/repositories`
 - add **lora oft** support, thanks @antis0007 and @ai-casanova
-- **upscalers compile** option, thanks @disty0
+- **upscalers**
+  - **compile compile** option, thanks @disty0
+  - **chainner** add high quality models from [Helaman](https://openmodeldb.info/users/helaman)
+  - **chainner** switch to `torchvision.transforms` for all image decode operations  
+- new option: *settings -> images -> keep incomplete*
+  can be used to skip vae decode on aborted/skipped/interrupted image generations
 
 
 **Themes**

README.md

@@ -69,7 +69,7 @@ Additional models will be added as they become available and there is public int
 - *Intel Arc* GPUs using **OneAPI** with *IPEX XPU* libraries on both *Windows and Linux*
 - Any GPU compatible with *DirectX* on *Windows* using **DirectML** libraries.  
   This includes support for AMD GPUs that are not supported by native ROCm libraries
-- Any GPU compatible with **OpenVINO** libraries on both *Windows and Linux*
+- Any GPU or device compatible with **OpenVINO** libraries on both *Windows and Linux*
 - *Apple M1/M2* on *OSX* using built-in support in Torch with **MPS** optimizations
 
 ## Install & Run

javascript/settings.js

@@ -77,7 +77,7 @@ function markIfModified(setting_name, value) {
   tab_nav_indicator.classList.toggle('saved', saved.size > 0);
   if (changed_items.size > 0) tab_nav_indicator.title += `click to reset ${changed_items.size} unapplied changes in this tab\n`;
   if (saved.size > 0) tab_nav_indicator.title += `${saved.size} custom values\n${unsaved.size} default values}`;
-  elem.scrollIntoView({ behavior: 'smooth', block: 'center' });
+  elem.scrollIntoView({ behavior: 'smooth', block: 'center' }); // TODO why is scroll happening on every change if all pages are visible?
 }
 
 onAfterUiUpdate(async () => {
@@ -120,18 +120,6 @@ onAfterUiUpdate(async () => {
   };
 });
 
-onOptionsChanged(() => {
-  const elem = gradioApp().getElementById('sd_checkpoint_hash');
-  const sd_checkpoint_hash = opts.sd_checkpoint_hash || '';
-  const shorthash = sd_checkpoint_hash.substring(0, 10);
-
-  if (elem && elem.textContent !== shorthash) {
-    elem.textContent = shorthash;
-    elem.title = sd_checkpoint_hash;
-    elem.href = `https://google.com/search?q=${sd_checkpoint_hash}`;
-  }
-});
-
 onOptionsChanged(() => {
   const setting_elems = gradioApp().querySelectorAll('#settings [id^="setting_"]');
   setting_elems.forEach((elem) => {

modules/dml/hijack/diffusers.py

@@ -20,7 +20,7 @@ def PNDMScheduler__get_prev_sample(self, sample: torch.FloatTensor, timestep, pr
     beta_prod_t = 1 - alpha_prod_t
     beta_prod_t_prev = 1 - alpha_prod_t_prev
 
-    if self.config.prediction_type == "v-prediction":
+    if self.config.prediction_type == "v_prediction":
         model_output = (alpha_prod_t**0.5) * model_output + (beta_prod_t**0.5) * sample
     elif self.config.prediction_type != "epsilon":
         raise ValueError(

modules/modelloader.py

@@ -183,7 +183,7 @@ def download_civit_model(model_url: str, model_name: str, model_path: str, model
     return f'CivitAI download: name={model_name} url={model_url} path={model_path}'
 
 
-def download_diffusers_model(hub_id: str, cache_dir: str = None, download_config: Dict[str, str] = None, token = None, variant = None, revision = None, mirror = None):
+def download_diffusers_model(hub_id: str, cache_dir: str = None, download_config: Dict[str, str] = None, token = None, variant = None, revision = None, mirror = None, custom_pipeline = None):
     if hub_id is None or len(hub_id) == 0:
         return None
     from diffusers import DiffusionPipeline
@@ -204,6 +204,8 @@ def download_diffusers_model(hub_id: str, cache_dir: str = None, download_config
         download_config["revision"] = revision
     if mirror is not None and len(mirror) > 0:
         download_config["mirror"] = mirror
+    if custom_pipeline is not None and len(custom_pipeline) > 0:
+        download_config["custom_pipeline"] = custom_pipeline
     shared.log.debug(f"Diffusers downloading: {hub_id} {download_config}")
     if token is not None and len(token) > 2:
         shared.log.debug(f"Diffusers authentication: {token}")

modules/progress.py

@@ -70,7 +70,7 @@ def progressapi(req: ProgressRequest):
     if shared.state.job_count > 0:
         progress += shared.state.job_no / shared.state.job_count
     if shared.state.sampling_steps > 0 and shared.state.job_count > 0:
-        progress += 1 / shared.state.job_count * shared.state.sampling_step / shared.state.sampling_steps
+        progress += 1 / (shared.state.job_count / 2 if shared.state.processing_has_refined_job_count else 1) * shared.state.sampling_step / shared.state.sampling_steps
     progress = min(progress, 1)
     elapsed_since_start = time.time() - shared.state.time_start
     predicted_duration = elapsed_since_start / progress if progress > 0 else None

modules/sd_models.py

@@ -178,7 +178,7 @@ def list_models():
                 shared.opts.data['sd_model_checkpoint'] = checkpoint_info.title
     elif shared.cmd_opts.ckpt != shared.default_sd_model_file and shared.cmd_opts.ckpt is not None:
         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')
+    shared.log.info(f'Available models: path="{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:
@@ -806,6 +806,9 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
     else:
         diffusers_load_config['variant'] = shared.opts.diffusers_model_load_variant
 
+    if shared.opts.diffusers_pipeline == 'Custom Diffusers Pipeline':
+        diffusers_load_config['custom_pipeline'] = shared.opts.custom_diffusers_pipeline
+
     if shared.opts.data.get('sd_model_checkpoint', '') == 'model.ckpt' or shared.opts.data.get('sd_model_checkpoint', '') == '':
         shared.opts.data['sd_model_checkpoint'] = "runwayml/stable-diffusion-v1-5"
 

modules/sd_vae.py

@@ -98,7 +98,7 @@ def refresh_vae_list():
                 vae_dict[name] = os.path.dirname(filepath)
             else:
                 vae_dict[name] = filepath
-    shared.log.info(f"Available VAEs: {vae_path} items={len(vae_dict)}")
+    shared.log.info(f'Available VAEs: path="{vae_path}" items={len(vae_dict)}')
     return vae_dict
 
 

modules/shared.py

@@ -337,6 +337,7 @@ options_templates.update(options_section(('diffusers', "Diffusers Settings"), {
     "diffusers_attention_slicing": OptionInfo(False, "Enable attention slicing"),
     "diffusers_model_load_variant": OptionInfo("default", "Diffusers model loading variant", gr.Radio, {"choices": ['default', 'fp32', 'fp16']}),
     "diffusers_vae_load_variant": OptionInfo("default", "Diffusers VAE loading variant", gr.Radio, {"choices": ['default', 'fp32', 'fp16']}),
+    "custom_diffusers_pipeline": OptionInfo('hf-internal-testing/diffusers-dummy-pipeline', 'Custom Diffusers pipeline to use'),
     "diffusers_lora_loader": OptionInfo("diffusers" if cmd_opts.use_openvino else "sequential apply", "Diffusers LoRA loading variant", gr.Radio, {"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"),
@@ -473,7 +474,7 @@ options_templates.update(options_section(('sampler-params', "Sampler Settings"),
     "schedulers_use_karras": OptionInfo(True, "Use Karras sigmas", gr.Checkbox, {"visible": False}),
     "schedulers_use_thresholding": OptionInfo(False, "Use dynamic thresholding", gr.Checkbox, {"visible": False}),
     "schedulers_use_loworder": OptionInfo(True, "Use simplified solvers in final steps", gr.Checkbox, {"visible": False}),
-    "schedulers_prediction_type": OptionInfo("default", "Override model prediction type", gr.Radio, {"choices": ['default', 'epsilon', 'sample', 'v-prediction'], "visible": False}),
+    "schedulers_prediction_type": OptionInfo("default", "Override model prediction type", gr.Radio, {"choices": ['default', 'epsilon', 'sample', 'v_prediction'], "visible": False}),
 
     # managed from ui.py for backend diffusers
     "schedulers_sep_diffusers": OptionInfo("<h2>Diffusers specific config</h2>", "", gr.HTML),

modules/shared_items.py

@@ -37,6 +37,7 @@ def get_pipelines():
         'Stable Diffusion XL Img2Img': getattr(diffusers, 'StableDiffusionXLImg2ImgPipeline', None),
         'Stable Diffusion XL Inpaint': getattr(diffusers, 'StableDiffusionXLInpaintPipeline', None),
         'Stable Diffusion XL Instruct': getattr(diffusers, 'StableDiffusionXLInstructPix2PixPipeline', None),
+        'Custom Diffusers Pipeline': getattr(diffusers, 'DiffusionPipeline', None),
         # 'Test': getattr(diffusers, 'TestPipeline', None),
         # 'Kandinsky V1', 'Kandinsky V2', 'DeepFloyd IF', 'Shap-E', 'Kandinsky V1 Img2Img', 'Kandinsky V2 Img2Img', 'DeepFloyd IF Img2Img', 'Shap-E Img2Img',
     }

modules/ui_models.py

@@ -203,9 +203,9 @@ def create_ui():
                 def hf_select(evt: gr.SelectData, data):
                     return data[evt.index[0]][0]
 
-                def hf_download_model(hub_id: str, token, variant, revision, mirror):
+                def hf_download_model(hub_id: str, token, variant, revision, mirror, custom_pipeline):
                     from modules.modelloader import download_diffusers_model
-                    download_diffusers_model(hub_id, cache_dir=opts.diffusers_dir, token=token, variant=variant, revision=revision, mirror=mirror)
+                    download_diffusers_model(hub_id, cache_dir=opts.diffusers_dir, token=token, variant=variant, revision=revision, mirror=mirror, custom_pipeline=custom_pipeline)
                     from modules.sd_models import list_models # pylint: disable=W0621
                     list_models()
                     log.info(f'Diffuser model downloaded: model="{hub_id}"')
@@ -227,6 +227,7 @@ def create_ui():
                     with gr.Row():
                         hf_token = gr.Textbox('', label = 'Huggingface token', placeholder='optional access token for private or gated models')
                         hf_mirror = gr.Textbox('', label = 'Huggingface mirror', placeholder='optional mirror site for downloads')
+                        hf_custom_pipeline = gr.Textbox('', label = 'Custom pipeline', placeholder='optional pipeline for downloads')
                 with gr.Column(scale=1):
                     gr.HTML('<br>')
                     hf_download_model_btn = gr.Button(value="Download model", variant='primary')
@@ -239,7 +240,7 @@ def create_ui():
                 hf_search_text.submit(fn=hf_search, inputs=[hf_search_text], outputs=[hf_results])
                 hf_search_btn.click(fn=hf_search, inputs=[hf_search_text], outputs=[hf_results])
                 hf_results.select(fn=hf_select, inputs=[hf_results], outputs=[hf_selected])
-                hf_download_model_btn.click(fn=hf_download_model, inputs=[hf_selected, hf_token, hf_variant, hf_revision, hf_mirror], outputs=[models_outcome])
+                hf_download_model_btn.click(fn=hf_download_model, inputs=[hf_selected, hf_token, hf_variant, hf_revision, hf_mirror, hf_custom_pipeline], outputs=[models_outcome])
 
             with gr.Tab(label="CivitAI"):
                 data = []