multiple patches

modules/hidiffusion/__init__.py

@@ -6,6 +6,9 @@ from modules.hidiffusion import hidiffusion
 
 
 def apply_hidiffusion(p, model_type):
+    if model_type not in ['sd', 'sdxl'] and p.hidiffusion:
+        shared.log.warning(f'HiDiffusion: class={shared.sd_model.__class__.__name__} not supported')
+        return
     remove_hidiffusion(p)
     if p.hidiffusion:
         t0 = time.time()

modules/processing_args.py

@@ -92,17 +92,20 @@ def set_pipeline_args(p, model, prompts: list, negative_prompts: list, prompts_2
 
     steps = kwargs.get("num_inference_steps", None) or len(getattr(p, 'timesteps', ['1']))
     if 'timesteps' in possible:
-        try:
-            timesteps = re.split(',| ', shared.opts.schedulers_timesteps)
-            timesteps = [int(x) for x in timesteps if x.isdigit()]
-            if len(timesteps) > 0:
-                args['timesteps'] = timesteps
-                p.steps = len(timesteps)
-                p.timesteps = timesteps
-                steps = p.steps
-                shared.log.debug(f'Sampler: steps={len(timesteps)} timesteps={timesteps}')
-        except Exception as e:
-            shared.log.error(f'Sampler timesteps: {e}')
+        if hasattr(model.scheduler, 'set_timesteps') and "timesteps" in set(inspect.signature(model.scheduler.set_timesteps).parameters.keys()):
+            try:
+                timesteps = re.split(',| ', shared.opts.schedulers_timesteps)
+                timesteps = [int(x) for x in timesteps if x.isdigit()]
+                if len(timesteps) > 0:
+                    args['timesteps'] = timesteps
+                    p.steps = len(timesteps)
+                    p.timesteps = timesteps
+                    steps = p.steps
+                    shared.log.debug(f'Sampler: steps={len(timesteps)} timesteps={timesteps}')
+            except Exception as e:
+                shared.log.error(f'Sampler timesteps: {e}')
+        else:
+            shared.log.warning(f'Sampler: sampler={model.scheduler.__class__.__name__} timesteps not supported')
     if shared.opts.prompt_attention != 'Fixed attention' and ('StableDiffusion' in model.__class__.__name__ or 'StableCascade' in model.__class__.__name__) and 'Onnx' not in model.__class__.__name__:
         try:
             prompt_parser_diffusers.encode_prompts(model, p, prompts, negative_prompts, steps=steps, clip_skip=clip_skip)

modules/processing_diffusers.py

@@ -85,6 +85,7 @@ def process_diffusers(p: processing.StableDiffusionProcessing):
 
     shared.sd_model = update_pipeline(shared.sd_model, p)
     shared.log.info(f'Base: class={shared.sd_model.__class__.__name__}')
+    update_sampler(p, shared.sd_model)
     base_args = set_pipeline_args(
         p=p,
         model=shared.sd_model,
@@ -102,7 +103,6 @@ def process_diffusers(p: processing.StableDiffusionProcessing):
         clip_skip=p.clip_skip,
         desc='Base',
     )
-    update_sampler(p, shared.sd_model)
     shared.state.sampling_steps = base_args.get('num_inference_steps', None) or p.steps
     p.extra_generation_params['Pipeline'] = shared.sd_model.__class__.__name__
     if shared.opts.scheduler_eta is not None and shared.opts.scheduler_eta > 0 and shared.opts.scheduler_eta < 1:
@@ -192,6 +192,7 @@ def process_diffusers(p: processing.StableDiffusionProcessing):
             sd_models.move_model(shared.sd_model, devices.device)
             orig_denoise = p.denoising_strength
             p.denoising_strength = getattr(p, 'hr_denoising_strength', p.denoising_strength)
+            update_sampler(p, shared.sd_model, second_pass=True)
             hires_args = set_pipeline_args(
                 p=p,
                 model=shared.sd_model,
@@ -209,7 +210,6 @@ def process_diffusers(p: processing.StableDiffusionProcessing):
                 strength=p.denoising_strength,
                 desc='Hires',
             )
-            update_sampler(p, shared.sd_model, second_pass=True)
             shared.state.job = 'HiRes'
             shared.state.sampling_steps = hires_args.get('num_inference_steps', None) or p.steps
             try:
@@ -257,6 +257,7 @@ def process_diffusers(p: processing.StableDiffusionProcessing):
             if hasattr(p, 'task_args') and p.task_args.get('image', None) is not None and output is not None: # replace input with output so it can be used by hires/refine
                 p.task_args['image'] = image
             shared.log.info(f'Refiner: class={shared.sd_refiner.__class__.__name__}')
+            update_sampler(p, shared.sd_refiner, second_pass=True)
             refiner_args = set_pipeline_args(
                 p=p,
                 model=shared.sd_refiner,
@@ -275,7 +276,6 @@ def process_diffusers(p: processing.StableDiffusionProcessing):
                 clip_skip=p.clip_skip,
                 desc='Refiner',
             )
-            update_sampler(p, shared.sd_refiner, second_pass=True)
             shared.state.sampling_steps = refiner_args.get('num_inference_steps', None) or p.steps
             try:
                 if 'requires_aesthetics_score' in shared.sd_refiner.config: # sdxl-model needs false and sdxl-refiner needs true

modules/sd_models.py

@@ -610,6 +610,8 @@ def detect_pipeline(f: str, op: str = 'model', warning=True):
             if 'stable-cascade' in f.lower() or 'stablecascade' in f.lower() or 'wuerstchen3' in f.lower():
                 if shared.backend == shared.Backend.ORIGINAL:
                     warn(f'Model detected as Stable Cascade model, but attempting to load using backend=original: {op}={f} size={size} MB')
+                if devices.dtype == torch.float16:
+                    warn('Stable Cascade does not support Float16')
                 guess = 'Stable Cascade'
             if 'pixart_sigma' in f.lower():
                 if shared.backend == shared.Backend.ORIGINAL:

modules/sd_samplers_diffusers.py

@@ -1,4 +1,5 @@
 import os
+import re
 import inspect
 from modules import shared
 from modules import sd_samplers_common
@@ -128,7 +129,9 @@ class DiffusionSampler:
         if shared.opts.schedulers_beta_schedule != 'default':
             self.config['beta_schedule'] = shared.opts.schedulers_beta_schedule
         if 'use_karras_sigmas' in self.config:
-            self.config['use_karras_sigmas'] = shared.opts.schedulers_use_karras
+            timesteps = re.split(',| ', shared.opts.schedulers_timesteps)
+            timesteps = [int(x) for x in timesteps if x.isdigit()]
+            self.config['use_karras_sigmas'] = shared.opts.schedulers_use_karras if len(timesteps) == 0 else False
         if 'thresholding' in self.config:
             self.config['thresholding'] = shared.opts.schedulers_use_thresholding
         if 'lower_order_final' in self.config: