fix: dead hasattr-on-dict path, double video-vae normalization, undefined self.device, hyimage if/elif, attention-mask attr typo, wrong face logged

Co-Authored-By: Claude <noreply@anthropic.com>

modules/control/processor.py

@@ -240,7 +240,7 @@ def preprocess_image(
             if 'strength' in possible:
                 p.task_args['strength'] = p.denoising_strength
             p.init_images = [init_image] * len(active_model)
-        if hasattr(shared.sd_model, 'controlnet') and hasattr(p.task_args, 'control_image') and len(p.task_args['control_image']) > 1 and (shared.sd_model.__class__.__name__ == 'StableDiffusionXLControlNetUnionPipeline'): # special case for controlnet-union
+        if hasattr(shared.sd_model, 'controlnet') and 'control_image' in p.task_args and len(p.task_args['control_image']) > 1 and (shared.sd_model.__class__.__name__ == 'StableDiffusionXLControlNetUnionPipeline'): # special case for controlnet-union
             p.task_args['control_image'] = [[x] for x in p.task_args['control_image']]
             p.task_args['control_mode'] = [[x] for x in p.task_args['control_mode']]
 

modules/face/instantid.py

@@ -35,7 +35,7 @@ def instant_id(p: processing.StableDiffusionProcessing, app, source_images, stre
         face = sorted(faces, key=lambda x:(x['bbox'][2]-x['bbox'][0])*x['bbox'][3]-x['bbox'][1])[-1]  # only use the maximum face
         face_embeds.append(torch.from_numpy(face['embedding']))
         face_images.append(draw_kps(source_image, face['kps']))
-        p.extra_generation_params[f"InstantID {i+1}"] = f'{faces[0].det_score:.2f} {"female" if faces[0].gender==0 else "male"} {faces[0].age}y'
+        p.extra_generation_params[f"InstantID {i+1}"] = f'{face.det_score:.2f} {"female" if face.gender==0 else "male"} {face.age}y'
         log.debug(f'InstantID face: score={face.det_score:.2f} gender={"female" if face.gender==0 else "male"} age={face.age} bbox={face.bbox}')
 
     log.debug(f'InstantID loading: model={REPO_ID}')

modules/postprocess/seedvr_model.py

@@ -27,6 +27,7 @@ class UpscalerSeedVR(Upscaler):
         ]
         self.model = None
         self.model_loaded = None
+        self.device = devices.device
 
     def load_model(self, path: str):
         model_name = MODELS_MAP.get(path, None)

modules/processing_class.py

@@ -633,7 +633,7 @@ class StableDiffusionProcessing:
         self.negative_embeds = []
         self.negative_pooleds = []
         self.prompt_attention_masks = []
-        self.negative_prompt_attention_mask = []
+        self.negative_prompt_attention_masks = []
         self.xyz = xyz
         self.abort = False
 

modules/video_models/video_vae.py

@@ -50,5 +50,4 @@ def vae_decode_tiny(latents):
     vae = vae.to(device=devices.device, dtype=devices.dtype)
     latents = latents.transpose(1, 2).to(device=devices.device, dtype=devices.dtype)
     images = vae.decode_video(latents, parallel=False).transpose(1, 2).mul_(2).sub_(1)
-    images = images.transpose(1, 2).mul_(2).sub_(1)
     return (images, None)

pipelines/model_hyimage.py

@@ -107,9 +107,9 @@ class HunyuanImage3Wrapper(torch.nn.Module):
 
         if height is None and width is None:
             image_size = "auto"
-        if height is None:
+        elif height is None:
             image_size = (width, width)
-        if width is None:
+        elif width is None:
             image_size = (height, height)
         else:
             image_size = (height, width)