Merge branch 'vladmandic:dev' into dev

.github/workflows/lint.yaml

@@ -4,6 +4,9 @@ on:
   - push
   - pull_request
 
+permissions:
+  contents: read
+
 jobs:
   lint:
     runs-on: ubuntu-24.04

CHANGELOG.md

@@ -1,8 +1,8 @@
 # Change Log for SD.Next
 
-## Update for 2026-06-14
+## Update for 2026-06-16
 
-### Highlights for 2026-06-14
+### Highlights for 2026-06-16
 
 *What's New?*
 - **Ideogram-4** released, Microsoft joins the game with **Lens** and **Anima** made it to release version
@@ -16,9 +16,11 @@ And we have new [Home page](https://vladmandic.github.io/sdnext/) with heavily u
 Plus continued work on modernization of codebase: UI is now fully TypeScript based
 And we have a new modular LoRA loader, new native Transformers loader and improved 3rd party finetunes support!  
 
-[ReadMe](https://github.com/vladmandic/automatic/blob/master/README.md) | [ChangeLog](https://github.com/vladmandic/automatic/blob/master/CHANGELOG.md) | [Docs](https://vladmandic.github.io/sdnext-docs/) | [WiKi](https://github.com/vladmandic/automatic/wiki) | [Discord](https://discord.com/invite/sd-next-federal-batch-inspectors-1101998836328697867) | [Sponsor](https://github.com/sponsors/vladmandic)  
+*Note*: This is a major update due to sheer size of the changes: over 400 commits!
 
-### Details for 2026-06-14
+[Home](https://vladmandic.github.io/sdnext/) | [ChangeLog](https://github.com/vladmandic/automatic/blob/master/CHANGELOG.md) | [Docs](https://vladmandic.github.io/sdnext-docs/) | [Discord](https://discord.com/invite/sd-next-federal-batch-inspectors-1101998836328697867) | [Sponsor](https://github.com/sponsors/vladmandic)  
+
+### Details for 2026-06-16
 
 - **Models**
   - [CircleStone Anima 1.0](https://huggingface.co/circlestone-labs/Anima) in *Base* and *Turbo* (distilled) variants  
@@ -150,8 +152,21 @@ And we have a new modular LoRA loader, new native Transformers loader and improv
   - `samplers` ui sigma methods
   - `xpu` generator on non-cpu
   - `compel` compatibility with *transformers==5*
+  - `gallery` open folder
+  - `seedvr` unload after upscale
+  - `tinyvae` with anima
   - `mixture-tiling` fix for non-square images, thanks @QualiaRain
   - `prompts-from-file` fix metadata handling, thanks @QualiaRain
+  - `hypertile` correct width/height assignment, thanks @QualiaRain
+  - custom allowed-paths, thanks @QualiaRain
+  - bias dtype, thanks @QualiaRain
+  - `ipadapter` mask accumulation, thanks @QualiaRain
+  - `lora` no-lora check, thanks @QualiaRain
+  - control `video` processing, thanks @QualiaRain
+  - `freescale` correct width/height assignment, thanks @QualiaRain
+  - noise `lerp` inversion, thanks @QualiaRain
+  - additional safety checks, thanks @QualiaRain
+  - `remote vae` shadowing, thanks @QualiaRain
 
 ## Update for 2026-05-13
 

modules/api/api.py

@@ -17,12 +17,12 @@ class Api:
         self.credentials = {}
         if shared.cmd_opts.auth:
             for auth in shared.cmd_opts.auth.split(","):
-                user, password = auth.split(":")
+                user, password = auth.split(":", 1)
                 self.credentials[user.replace('"', '').strip()] = password.replace('"', '').strip()
         if shared.cmd_opts.auth_file:
             with open(shared.cmd_opts.auth_file, encoding="utf8") as file:
                 for line in file.readlines():
-                    user, password = line.split(":")
+                    user, password = line.split(":", 1)
                     self.credentials[user.replace('"', '').strip()] = password.replace('"', '').strip()
         self.router = APIRouter()
         if shared.cmd_opts.docs:

modules/api/gallery.py

@@ -3,7 +3,6 @@ import os
 import time
 import base64
 from urllib.parse import quote, unquote
-from fastapi import FastAPI
 from fastapi.responses import JSONResponse
 from starlette.websockets import WebSocket, WebSocketState
 from pydantic import BaseModel, Field # pylint: disable=no-name-in-module
@@ -173,7 +172,7 @@ def register_api(api): # register api
                 unique_folders.append(f)
                 if shared.demo is not None and path not in shared.demo.allowed_paths:
                     debug(f'Browser folders allow: {path}')
-                    shared.demo.allowed_paths.append(quote(path))
+                    shared.demo.allowed_paths.append(path)
         debug(f'Browser folders: {unique_folders}')
         return JSONResponse(content=unique_folders)
 

modules/api/generate.py

@@ -70,6 +70,7 @@ class APIGenerate:
             p.ip_adapter_starts = []
             p.ip_adapter_ends = []
             p.ip_adapter_images = []
+            p.ip_adapter_masks = []
             for ipadapter in request.ip_adapter:
                 if not ipadapter.images or len(ipadapter.images) == 0:
                     continue
@@ -79,7 +80,6 @@ class APIGenerate:
                 p.ip_adapter_starts.append(ipadapter.start)
                 p.ip_adapter_ends.append(ipadapter.end)
                 p.ip_adapter_images.append([helpers.decode_base64_to_image(x) for x in ipadapter.images])
-                p.ip_adapter_masks = []
                 if ipadapter.masks:
                     p.ip_adapter_masks.append([helpers.decode_base64_to_image(x) for x in ipadapter.masks])
             del request.ip_adapter

modules/api/process.py

@@ -76,7 +76,7 @@ class APIProcess:
         if processor is None or processor.processor_id != req.model:
             with self.queue_lock:
                 processor = processors.Processor(req.model)
-        for k, v in req.params.items():
+        for k, v in (req.params or {}).items():
             if k not in processors.config[processor.processor_id]['params']:
                 return JSONResponse(status_code=400, content={"error": f"Processor invalid parameter: id={req.model} {k}={v}"})
         jobid = shared.state.begin('API-PRE', api=True)
@@ -102,7 +102,7 @@ class APIProcess:
             return JSONResponse(status_code=400, content={"error": f"Mask type not found: id={req.type}"})
         image = decode_base64_to_image(req.image)
         mask = decode_base64_to_image(req.mask) if req.mask else None
-        for k, v in req.params.items():
+        for k, v in (req.params or {}).items():
             if not hasattr(masking.opts, k):
                 return JSONResponse(status_code=400, content={"error": f"Mask invalid parameter: {k}={v}"})
             else:

modules/control/run.py

@@ -644,7 +644,6 @@ def control_run(state: str = '', # pylint: disable=keyword-arg-before-vararg
     debug_log(f'Control pipeline: class={pipe.__class__.__name__} args={vars(p)}')
     status = True
     frame = None
-    video = None
     output_filename = None
     index = 0
     frames = 0
@@ -697,7 +696,7 @@ def control_run(state: str = '', # pylint: disable=keyword-arg-before-vararg
                     inputs = [Image.fromarray(frame)] # cv2 to pil
                 for i, input_image in enumerate(inputs): # loop per-input, but with early-break
                     if pipe is None: # pipe may have been reset externally
-                        if video is None:
+                        if cap is None:
                             break # non-video: pipeline was consumed, no need to re-process remaining inputs
                         pipe = set_pipe(p, has_models, unit_type, selected_models, active_model, active_strength, active_units, control_conditioning, control_guidance_start, control_guidance_end, inits)
                         debug_log(f'Control pipeline reinit: class={pipe.__class__.__name__}')
@@ -744,7 +743,7 @@ def control_run(state: str = '', # pylint: disable=keyword-arg-before-vararg
                         init_image = inits[i % len(inits)]
                     else:
                         init_image = None
-                    if video is not None and index % (video_skip_frames + 1) != 0:
+                    if cap is not None and index % (video_skip_frames + 1) != 0:
                         index += 1
                         continue
                     index += 1
@@ -806,13 +805,13 @@ def control_run(state: str = '', # pylint: disable=keyword-arg-before-vararg
                                 if processed_image is not None and isinstance(processed_image, Image.Image):
                                     output_images.append(processed_image)
 
-                            if is_generator and frame is not None and video is not None:
+                            if is_generator and frame is not None and cap is not None:
                                 image_txt = f'{output_image.width}x{output_image.height}' if output_image is not None else 'None'
                                 msg = f'Control output | {index} of {frames} skip {video_skip_frames} | Frame {image_txt}'
                                 yield (output_image, blended_image, msg) # result is control_output, proces_output
 
-                if video is not None and frame is not None:
-                    status, frame = video.read()
+                if cap is not None and frame is not None:
+                    status, frame = cap.read()
                     if status:
                         frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
                     debug_log(f'Control: video frame={index} frames={frames} status={status} skip={index % (video_skip_frames + 1)} progress={index/frames:.2f}')

modules/control/units/controlnet.py

@@ -179,7 +179,7 @@ def api_list_models(model_type: str | None = None):
         model_list += list(predefined_qwen)
     if model_type == 'hunyuandit' or model_type == 'all':
         model_list += list(predefined_hunyuandit)
-    if model_type == 'zimage':
+    if model_type == 'zimage' or model_type == 'all':
         model_list += list(predefined_zimage)
     model_list += sorted(find_models())
     return model_list

modules/control/units/lite_model.py

@@ -171,7 +171,7 @@ class ControlNetLLLite(torch.nn.Module): # pylint: disable=abstract-method
                 mapped_block, mapped_number = map_down_lllite_to_unet[int(block)]
                 b = model.down_blocks[mapped_block].attentions[int(mapped_number)].transformer_blocks[int(block_number)]
             elif root == 'output':
-                pass # not implemented
+                continue # not implemented
             else:
                 b = model.mid_block.attentions[0].transformer_blocks[int(block_number)]
             b = getattr(b, attn_name, None)

modules/lora/lora_extract.py

@@ -122,7 +122,7 @@ def make_lora(fn, maxrank, auto_rank, rank_ratio, modules, overwrite):
         log.warning(msg)
         yield msg
         return
-    if loaded_lora() == "":
+    if not loaded_lora():
         msg = "LoRA extract: no LoRA detected"
         log.warning(msg)
         yield msg
@@ -141,8 +141,7 @@ def make_lora(fn, maxrank, auto_rank, rank_ratio, modules, overwrite):
     with rp.Progress(rp.TextColumn('[cyan]LoRA extract'), rp.BarColumn(), rp.TaskProgressColumn(), rp.TimeRemainingColumn(), rp.TimeElapsedColumn(), rp.TextColumn('[cyan]{task.description}'), console=console) as progress:
 
         if 'te' in modules and getattr(shared.sd_model, 'text_encoder', None) is not None:
-            modules = shared.sd_model.text_encoder.named_modules()
-            task = progress.add_task(description="te1 decompose", total=len(list(modules)))
+            task = progress.add_task(description="te1 decompose", total=len(list(shared.sd_model.text_encoder.named_modules())))
             for name, module in shared.sd_model.text_encoder.named_modules():
                 progress.update(task, advance=1)
                 weights_backup = getattr(module, "network_weights_backup", None)
@@ -157,8 +156,7 @@ def make_lora(fn, maxrank, auto_rank, rank_ratio, modules, overwrite):
         t1 = time.time()
 
         if 'te' in modules and getattr(shared.sd_model, 'text_encoder_2', None) is not None:
-            modules = shared.sd_model.text_encoder_2.named_modules()
-            task = progress.add_task(description="te2 decompose", total=len(list(modules)))
+            task = progress.add_task(description="te2 decompose", total=len(list(shared.sd_model.text_encoder_2.named_modules())))
             for name, module in shared.sd_model.text_encoder_2.named_modules():
                 progress.update(task, advance=1)
                 weights_backup = getattr(module, "network_weights_backup", None)
@@ -172,8 +170,7 @@ def make_lora(fn, maxrank, auto_rank, rank_ratio, modules, overwrite):
         t2 = time.time()
 
         if 'unet' in modules and getattr(shared.sd_model, 'unet', None) is not None:
-            modules = shared.sd_model.unet.named_modules()
-            task = progress.add_task(description="unet decompose", total=len(list(modules)))
+            task = progress.add_task(description="unet decompose", total=len(list(shared.sd_model.unet.named_modules())))
             for name, module in shared.sd_model.unet.named_modules():
                 progress.update(task, advance=1)
                 weights_backup = getattr(module, "network_weights_backup", None)

modules/lora/network.py

@@ -166,6 +166,7 @@ class NetworkModule:
         self.network_key = weights.network_key
         self.sd_key = weights.sd_key
         self.sd_module = weights.sd_module
+        self.shape = None
         if hasattr(self.sd_module, 'weight'):
             if hasattr(self.sd_module, "sdnq_dequantizer"):
                 self.shape = self.sd_module.sdnq_dequantizer.original_shape
@@ -176,7 +177,7 @@ class NetworkModule:
         self.alpha = weights.w["alpha"].item() if "alpha" in weights.w else None
         self.scale = weights.w["scale"].item() if "scale" in weights.w else None
         self.dora_scale = weights.w.get("dora_scale", None)
-        self.dora_norm_dims = len(self.shape) - 1
+        self.dora_norm_dims = (len(self.shape) - 1) if self.shape is not None else None
 
     def multiplier(self):
         unet_multiplier = 3 * [self.network.unet_multiplier] if not isinstance(self.network.unet_multiplier, list) else self.network.unet_multiplier

modules/openai/test.py

@@ -20,7 +20,13 @@ model = AutoModelForCausalLM.from_pretrained(
     dtype=torch.bfloat16,
     trust_remote_code=True,
     attn_implementation="sdpa",
+    # attn_implementation="eager",
+    # attn_implementation="flash_attention_2",
 )
+model.config.use_flash_attention = True
+
+logger.log.info("OpenAI: eval model...")
+model.eval()
 tokenizer = AutoTokenizer.from_pretrained(
     "Qwen/Qwen3-0.6B",
     trust_remote_code=True

modules/postprocess/seedvr_model.py

@@ -47,7 +47,10 @@ class UpscalerSeedVR(Upscaler):
             self.model.dit.dtype = devices.dtype
             self.model.vae_encode = self.vae_encode
             self.model.vae_decode = self.vae_decode
-            self.model.model_step = generation.generation_step
+            # Patch generation_loop's generation_step() with our wrapper; stash the original once
+            # so reloads don't re-wrap the wrapper itself (infinite recursion).
+            if not hasattr(generation, "generation_step_original"):
+                generation.generation_step_original = generation.generation_step
             generation.generation_step = self.model_step
             self.model._internal_dict = {
                 'dit': self.model.dit,
@@ -119,6 +122,7 @@ class UpscalerSeedVR(Upscaler):
         return samples
 
     def model_step(self, *args, **kwargs):
+        from modules.seedvr.src.core import generation
         from modules.seedvr.src.optimization import memory_manager
         self.model.vae = self.model.vae.to(device="cpu")
         self.model.dit = self.model.dit.to(device=self.device)
@@ -126,7 +130,7 @@ class UpscalerSeedVR(Upscaler):
         log.debug(f'Upscaler inference: args={len(args)} kwargs={list(kwargs.keys())}')
         memory_manager.preinitialize_rope_cache(self.model)
         with devices.inference_context():
-            result = self.model.model_step(*args, **kwargs)
+            result = generation.generation_step_original(*args, **kwargs)
         self.model.dit = self.model.dit.to(device="cpu")
         devices.torch_gc()
         return result

modules/processing_vae.py

@@ -246,13 +246,21 @@ def vae_postprocess(tensor, model, output_type='np'):
             if hasattr(model, 'video_processor'):
                 if tensor.ndim == 6 and tensor.shape[1] == 1:
                     tensor = tensor.squeeze(0)
-                images = model.video_processor.postprocess_video(tensor, output_type='pil')
+                try:
+                    images = model.video_processor.postprocess_video(tensor, output_type='pil')
+                except Exception as e:
+                    log.warning(f'VAE postprocess: type=video {e}')
+                    images = tensor
                 if isinstance(images, list) and len(images) > 0 and isinstance(images[0], list):
                     images = [frame for batch in images for frame in batch]
             elif hasattr(model, 'image_processor'):
                 if tensor.ndim == 5 and tensor.shape[1] == 3: # Qwen Image
                     tensor = tensor[:, :, 0]
-                images = model.image_processor.postprocess(tensor, output_type=output_type)
+                try:
+                    images = model.image_processor.postprocess(tensor, output_type=output_type)
+                except Exception as e:
+                    log.warning(f'VAE postprocess: type=image {e}')
+                    images = tensor
             elif hasattr(model, "vqgan"):
                 images = tensor.permute(0, 2, 3, 1).cpu().float().numpy()
                 if output_type == "pil":
@@ -263,6 +271,12 @@ def vae_postprocess(tensor, model, output_type='np'):
                 if tensor.ndim == 5 and tensor.shape[1] == 3: # Qwen Image
                     tensor = tensor[:, :, 0]
                 images = model.image_processor.postprocess(tensor, output_type=output_type)
+            if torch.is_tensor(images): # failed to postprocess, do naive conversion
+                images = images.permute(0, 2, 3, 1).cpu().float().numpy()
+                if images.min() < 0 or images.max() > 1:
+                    images = (images - images.min()) / (images.max() - images.min()) # naive normalization
+                if output_type == "pil":
+                    images = model.numpy_to_pil(images)
         else:
             images = tensor if isinstance(tensor, list) or isinstance(tensor, np.ndarray) else [tensor]
     except Exception as e:

modules/sd_detect.py

@@ -187,6 +187,7 @@ def guess_by_diffusers(fn, current_guess):
         cls = index.get('_class_name', None)
         if isinstance(cls, list):
             cls = cls[-1]
+        pipeline = None
         if cls is not None:
             pipeline = getattr(diffusers, cls, None)
             if pipeline is None:

modules/sd_hijack_accelerate.py

@@ -84,7 +84,7 @@ def torch_conv_forward(self, input, weight, bias): # pylint: disable=redefined-b
     if self.padding_mode != 'zeros':
         return F.conv2d(F.pad(input, self._reversed_padding_repeated_twice, mode=self.padding_mode), weight, bias, self.stride, _pair(0), self.dilation, self.groups) # pylint: disable=protected-access
     if weight.dtype != bias.dtype:
-        bias.to(weight.dtype)
+        bias = bias.to(weight.dtype)
     return F.conv2d(input, weight, bias, self.stride, self.padding, self.dilation, self.groups)
 
 def hijack_torch_conv():

modules/sd_hijack_hypertile.py

@@ -259,8 +259,8 @@ def set_resolution(p, hr=False):
     if hr:
         x = getattr(p, 'hr_upscale_to_x', 0)
         y = getattr(p, 'hr_upscale_to_y', 0)
-        width = y if y > 0 else p.width
-        height = x if x > 0 else p.height
+        width = x if x > 0 else p.width
+        height = y if y > 0 else p.height
     else:
         width = p.width
         height = p.height

modules/sd_hijack_te.py

@@ -8,7 +8,7 @@ def hijack_encode_prompt(*args, **kwargs):
     jobid = shared.state.begin('TE Encode')
     t0 = time.time()
     if 'max_sequence_length' in kwargs and kwargs['max_sequence_length'] is not None:
-        kwargs['max_sequence_length'] = max(kwargs['max_sequence_length'], os.environ.get('MAX_SEQUENCE_LENGTH', 256))
+        kwargs['max_sequence_length'] = max(kwargs['max_sequence_length'], int(os.environ.get('MAX_SEQUENCE_LENGTH', 256)))
     res = None
     try:
         args_copy = list(args)

modules/ui_common.py

@@ -264,6 +264,8 @@ def save_files(js_data, files, html_info, index):
 
 def open_folder(result_gallery, gallery_index = 0):
     try:
+        if gallery_index >= len(result_gallery):
+            gallery_index = 0
         folder = os.path.dirname(result_gallery[gallery_index]['name'])
     except Exception:
         folder = shared.opts.outdir_samples

modules/video_models/video_prompt.py

@@ -1,7 +1,7 @@
 from modules import shared, extra_networks, ui_video_vlm
 
 
-def prepare_prompt(p, init_image, prompt:str, vlm_enhance:bool, vlm_model:str, vlm_system_prompt:str):
+def prepare_prompts(p, init_image, prompt:str, vlm_enhance:bool, vlm_model:str, vlm_system_prompt:str):
     p.prompt = shared.prompt_styles.apply_styles_to_prompt(p.prompt, p.styles)
     p.negative_prompt = shared.prompt_styles.apply_negative_styles_to_prompt(p.negative_prompt, p.styles)
     shared.prompt_styles.apply_styles_to_extra(p)
@@ -18,4 +18,7 @@ def prepare_prompt(p, init_image, prompt:str, vlm_enhance:bool, vlm_model:str, v
     )
     if new_prompt is not None and len(new_prompt) > 0:
         prompt = new_prompt
-    return prompt
+
+    p.styles = []
+    p.task_args['prompt'] = p.prompt
+    p.task_args['negative_prompt'] = p.negative_prompt

modules/video_models/video_run.py

@@ -95,12 +95,11 @@ def generate(*args, **kwargs):
     shared.sd_model = sd_models.apply_balanced_offload(shared.sd_model)
     devices.torch_gc(force=True, reason='video')
 
-    prompt = video_prompt.prepare_prompt(p, init_image, prompt, vlm_enhance, vlm_model, vlm_system_prompt)
 
     # set args
+    video_prompt.prepare_prompts(p, init_image, prompt, vlm_enhance, vlm_model, vlm_system_prompt)
     processing.fix_seed(p)
     video_vae.set_vae_params(p)
-    video_utils.set_prompt(p)
     p.task_args['num_inference_steps'] = p.steps
     p.task_args['width'] = p.width
     p.task_args['height'] = p.height

modules/video_models/video_utils.py

@@ -30,15 +30,6 @@ def check_av():
     return av
 
 
-def set_prompt(p):
-    p.prompt = shared.prompt_styles.apply_styles_to_prompt(p.prompt, p.styles)
-    p.negative_prompt = shared.prompt_styles.apply_negative_styles_to_prompt(p.negative_prompt, p.styles)
-    shared.prompt_styles.apply_styles_to_extra(p)
-    p.styles = []
-    p.task_args['prompt'] = p.prompt
-    p.task_args['negative_prompt'] = p.negative_prompt
-
-
 def hijack_encode_image(*args, **kwargs):
     t0 = time.time()
     try: