Closed srcrist closed 8 months ago
i've modified logic, so upgrade and try again.
also added some debug statements, triggered by setting env variable SD_STEPS_DEBUG=true
note that using standard model as refiner is triggering different logic than using actual refiner model as standard model does not have denoise start/end.
and example with sdxl base + sdxl refiner would be that if you have base steps 10 and refiner start at 0.5, it will actually set steps to 20, but tell model to only run 0.5x of them and then pass unfinished results to refiner which means progress bar will only go to half before it stops - this is ideal workflow for refiner.
using normal model as refiner is basically using it as denoiser and looking at denoise strength, not refiner start.
so i'd need a bit more info here: which model is used as base, which model is used as refiner and screenshot of all second pass settings as denoise strength is also taken into account and its not shown here.
btw, new debug output looks like:
Steps: type=base input=20 output=40.0 refiner=True
it tells you what setting you passed, what was calculated and why.
closing as resolved for now, but if there are further issues, post here and i'll reopen.
Original 99 step issue resolved for SDXL refiner. Pursuant to our discussion on discord, here is where we're at with the SD1.5 "refiners":
The expected result should be something like the following:
Original
5 Refiner Steps
10 Refiner Steps
15 Refiner Steps
Instead, SDNext is presently producing the following:
Original
5 Refiner Steps
2023-11-09 18:15:28,768 | sd | DEBUG | processing_diffusers | Steps: type=refiner input=10 output=5.0 start=0.5 denoise=0.5
10 Refiner Steps
2023-11-09 18:15:42,303 | sd | DEBUG | processing_diffusers | Steps: type=refiner input=20 output=10.0 start=0.5 denoise=0.5
15 Refiner Steps
2023-11-09 18:34:12,121 | sd | DEBUG | processing_diffusers | Steps: type=refiner input=30 output=15.0 start=0.5 denoise=0.5
The change in the image is far more abrupt than I would expect, and the low-step generations are qualitatively bad as well.
specify models and settings for producing those examples, lets remove guesswork.
Models are Base: darkSushiMixMix_colorful [969673ad74]
and Refiner: realisticVisionV51_v51VAE [15012c538f]
The SDNext set above are admittedly a different set of models, so I did another set using the same base to eliminate variables. I'll include logs for the full generation details below.
Original
((masterpiece)),((highres)),((best quality)),((highly detailed)),((extremely detailed CG unity 8k wallpaper)), detailed face, illustration,Sharon,solo,looking at viewer,crossed arms,expressionless,closed mouth,outdoors,dusk,street,from side
Negative prompt: (human, skin:1.3), worst quality, low res, bad anatomy, deformed, (extra limbs, extra legs, bad hands), extra ears, (easynegative, by bad-artist-anime, bad-hands-5, bad_prompt_version2)
Steps: 30, Seed: 43534534, Sampler: KDPM2, CFG scale: 7, Size: 768x768, Parser: Full parser, Model: darkSushiMixMix_colorful, Model hash: 969673ad74, VAE: vae-ft-mse-840000-ema-pruned, Clip skip: 2, Backend: Diffusers, App: SD.Next, Version: baed872, Operations: txt2img, Pipeline: StableDiffusionPipeline, CFG rescale: 0.7, Embeddings: "bad_prompt_version2, easynegative, bad-hands-5, bad-artist-anime"
Template: ((masterpiece)),((highres)),((best quality)),((highly detailed)),((extremely detailed CG unity 8k wallpaper)), detailed face, illustration,Sharon,solo,looking at viewer,crossed arms,expressionless,closed mouth,outdoors,dusk,street,from side
Negative Template: (human, skin:1.3), worst quality, low res, bad anatomy, deformed, (extra limbs, extra legs, bad hands), extra ears, (easynegative, by bad-artist-anime, bad-hands-5, bad_prompt_version2)
5 Refiner Steps
((masterpiece)),((highres)),((best quality)),((highly detailed)),((extremely detailed CG unity 8k wallpaper)), detailed face, illustration,Sharon,solo,looking at viewer,crossed arms,expressionless,closed mouth,outdoors,dusk,street,from side
Negative prompt: (human, skin:1.3), worst quality, low res, bad anatomy, deformed, (extra limbs, extra legs, bad hands), extra ears, (easynegative, by bad-artist-anime, bad-hands-5, bad_prompt_version2)
Steps: 30, Seed: 43534534, Sampler: KDPM2, CFG scale: 7, Size: 768x768, Parser: Full parser, Model: darkSushiMixMix_colorful, Model hash: 969673ad74, VAE: vae-ft-mse-840000-ema-pruned, Clip skip: 2, Backend: Diffusers, App: SD.Next, Version: baed872, Operations: refine; txt2img, Second pass: True, Refiner: realisticVisionV51_v51VAE, Image CFG scale: 6, Refiner steps: 10, Refiner start: 0.5, Hires steps: 30, Latent sampler: Heun, CFG rescale: 0.7, Pipeline: StableDiffusionPipeline, Embeddings: "bad_prompt_version2, easynegative, bad-hands-5, bad-artist-anime"
Template: ((masterpiece)),((highres)),((best quality)),((highly detailed)),((extremely detailed CG unity 8k wallpaper)), detailed face, illustration,Sharon,solo,looking at viewer,crossed arms,expressionless,closed mouth,outdoors,dusk,street,from side
Negative Template: (human, skin:1.3), worst quality, low res, bad anatomy, deformed, (extra limbs, extra legs, bad hands), extra ears, (easynegative, by bad-artist-anime, bad-hands-5, bad_prompt_version2)
10 Refiner Steps
((masterpiece)),((highres)),((best quality)),((highly detailed)),((extremely detailed CG unity 8k wallpaper)), detailed face, illustration,Sharon,solo,looking at viewer,crossed arms,expressionless,closed mouth,outdoors,dusk,street,from side
Negative prompt: (human, skin:1.3), worst quality, low res, bad anatomy, deformed, (extra limbs, extra legs, bad hands), extra ears, (easynegative, by bad-artist-anime, bad-hands-5, bad_prompt_version2)
Steps: 30, Seed: 43534534, Sampler: KDPM2, CFG scale: 7, Size: 768x768, Parser: Full parser, Model: darkSushiMixMix_colorful, Model hash: 969673ad74, VAE: vae-ft-mse-840000-ema-pruned, Clip skip: 2, Backend: Diffusers, App: SD.Next, Version: baed872, Operations: refine; txt2img, Second pass: True, Refiner: realisticVisionV51_v51VAE, Image CFG scale: 6, Refiner steps: 20, Refiner start: 0.5, Hires steps: 30, Latent sampler: Heun, CFG rescale: 0.7, Pipeline: StableDiffusionPipeline, Embeddings: "bad_prompt_version2, easynegative, bad-hands-5, bad-artist-anime"
Template: ((masterpiece)),((highres)),((best quality)),((highly detailed)),((extremely detailed CG unity 8k wallpaper)), detailed face, illustration,Sharon,solo,looking at viewer,crossed arms,expressionless,closed mouth,outdoors,dusk,street,from side
Negative Template: (human, skin:1.3), worst quality, low res, bad anatomy, deformed, (extra limbs, extra legs, bad hands), extra ears, (easynegative, by bad-artist-anime, bad-hands-5, bad_prompt_version2)
15 Refiner Steps
((masterpiece)),((highres)),((best quality)),((highly detailed)),((extremely detailed CG unity 8k wallpaper)), detailed face, illustration,Sharon,solo,looking at viewer,crossed arms,expressionless,closed mouth,outdoors,dusk,street,from side
Negative prompt: (human, skin:1.3), worst quality, low res, bad anatomy, deformed, (extra limbs, extra legs, bad hands), extra ears, (easynegative, by bad-artist-anime, bad-hands-5, bad_prompt_version2)
Steps: 30, Seed: 43534534, Sampler: KDPM2, CFG scale: 7, Size: 768x768, Parser: Full parser, Model: darkSushiMixMix_colorful, Model hash: 969673ad74, VAE: vae-ft-mse-840000-ema-pruned, Clip skip: 2, Backend: Diffusers, App: SD.Next, Version: baed872, Operations: refine; txt2img, Second pass: True, Refiner: realisticVisionV51_v51VAE, Image CFG scale: 6, Refiner steps: 30, Refiner start: 0.5, Hires steps: 30, Latent sampler: Heun, CFG rescale: 0.7, Pipeline: StableDiffusionPipeline, Embeddings: "bad_prompt_version2, easynegative, bad-hands-5, bad-artist-anime"
Template: ((masterpiece)),((highres)),((best quality)),((highly detailed)),((extremely detailed CG unity 8k wallpaper)), detailed face, illustration,Sharon,solo,looking at viewer,crossed arms,expressionless,closed mouth,outdoors,dusk,street,from side
Negative Template: (human, skin:1.3), worst quality, low res, bad anatomy, deformed, (extra limbs, extra legs, bad hands), extra ears, (easynegative, by bad-artist-anime, bad-hands-5, bad_prompt_version2)
Debug logs for these 4 images:
2023-11-09 20:42:09,588 | sd | DEBUG | txt2img | txt2img: id_task=task(3nmtz7qhbusimlk)|prompt=((masterpiece)),((highres)),((best quality)),((highly detailed)),((extremely detailed CG unity 8k wallpaper)), detailed face, illustration,Sharon,solo,looking at viewer,crossed arms,expressionless,closed mouth,outdoors,dusk,street,from side|negative_prompt=(human, skin:1.3), worst quality, low res, bad anatomy, deformed, (extra limbs, extra legs, bad hands), extra ears, (easynegative, by bad-artist-anime, bad-hands-5, bad_prompt_version2)|prompt_styles=[]|steps=30|sampler_index=7|latent_index=14|full_quality=True|restore_faces=False|tiling=False|n_iter=1|batch_size=1|cfg_scale=7|clip_skip=2|seed=43534534.0|subseed=-1.0|subseed_strength=0|seed_resize_from_h=0|seed_resize_from_w=0||height=768|width=768|enable_hr=False|denoising_strength=0.5|hr_scale=2|hr_upscaler=None|hr_force=True|hr_second_pass_steps=30|hr_resize_x=0|hr_resize_y=0|image_cfg_scale=6|diffusers_guidance_rescale=0.7|refiner_steps=10|refiner_start=0.5|refiner_prompt=|refiner_negative=|override_settings_texts=[]
2023-11-09 20:42:09,605 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=all config={'num_train_timesteps': 1000, 'beta_start': 0.0001, 'beta_end': 0.02, 'beta_schedule': 'linear', 'prediction_type': 'epsilon'}
2023-11-09 20:42:09,607 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=scheduler config={'num_train_timesteps': 1000, 'beta_start': 0.0001, 'beta_end': 0.02, 'beta_schedule': 'linear', 'prediction_type': 'epsilon', 'steps_offset': 0}
2023-11-09 20:42:09,610 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=model config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'steps_offset': 1}
2023-11-09 20:42:09,613 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=user config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'steps_offset': 1}
2023-11-09 20:42:09,811 | sd | DEBUG | sd_samplers | Sampler: sampler="KDPM2" config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'steps_offset': 1}
2023-11-09 20:42:10,107 | sd | DEBUG | processing_diffusers | Steps: type=base input=30 output=30 refiner=False
2023-11-09 20:42:10,197 | sd | DEBUG | processing_diffusers | Diffuser pipeline: StableDiffusionPipeline task=DiffusersTaskType.TEXT_2_IMAGE set={'prompt_embeds': torch.Size([1, 77, 768]), 'negative_prompt_embeds': torch.Size([1, 77, 768]), 'guidance_scale': 7, 'generator': device(type='cpu'), 'output_type': 'latent', 'num_inference_steps': 30, 'eta': 1.0, 'guidance_rescale': 0.7, 'height': 768, 'width': 768, 'parser': 'Full parser'}
2023-11-09 20:42:14,387 | sd | DEBUG | processing_diffusers | Moving to CPU: model=UNet
2023-11-09 20:42:16,258 | sd | DEBUG | processing_diffusers | VAE decode: name=vae-ft-mse-840000-ema-pruned.safetensors dtype=torch.float16 upcast=False images=1 latents=torch.Size([1, 4, 96, 96]) time=1.87
2023-11-09 20:42:16,272 | sd | DEBUG | images | Saving: image="D:\StableDiffusion\outputs\text\2023-11-09\00068-20231109204216-b285e3fb-7108-4f3c-92a2-343dacc62a16.png" type=PNG size=768x768
2023-11-09 20:42:16,414 | sd | INFO | processing | Processed: images=1 time=6.81 its=4.41 memory={'ram': {'used': 6.08, 'total': 127.77}, 'gpu': {'used': 3.72, 'total': 23.99}, 'retries': 0, 'oom': 0}
2023-11-09 20:42:23,955 | sd | DEBUG | txt2img | txt2img: id_task=task(fn8lcwvhj47ts5v)|prompt=((masterpiece)),((highres)),((best quality)),((highly detailed)),((extremely detailed CG unity 8k wallpaper)), detailed face, illustration,Sharon,solo,looking at viewer,crossed arms,expressionless,closed mouth,outdoors,dusk,street,from side|negative_prompt=(human, skin:1.3), worst quality, low res, bad anatomy, deformed, (extra limbs, extra legs, bad hands), extra ears, (easynegative, by bad-artist-anime, bad-hands-5, bad_prompt_version2)|prompt_styles=[]|steps=30|sampler_index=7|latent_index=14|full_quality=True|restore_faces=False|tiling=False|n_iter=1|batch_size=1|cfg_scale=7|clip_skip=2|seed=43534534.0|subseed=-1.0|subseed_strength=0|seed_resize_from_h=0|seed_resize_from_w=0||height=768|width=768|enable_hr=True|denoising_strength=0.5|hr_scale=2|hr_upscaler=None|hr_force=True|hr_second_pass_steps=30|hr_resize_x=0|hr_resize_y=0|image_cfg_scale=6|diffusers_guidance_rescale=0.7|refiner_steps=10|refiner_start=0.5|refiner_prompt=|refiner_negative=|override_settings_texts=[]
2023-11-09 20:42:23,967 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=all config={'num_train_timesteps': 1000, 'beta_start': 0.0001, 'beta_end': 0.02, 'beta_schedule': 'linear', 'prediction_type': 'epsilon'}
2023-11-09 20:42:23,969 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=scheduler config={'num_train_timesteps': 1000, 'beta_start': 0.0001, 'beta_end': 0.02, 'beta_schedule': 'linear', 'prediction_type': 'epsilon', 'steps_offset': 0}
2023-11-09 20:42:23,972 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=model config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'steps_offset': 1}
2023-11-09 20:42:23,974 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=user config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'steps_offset': 1}
2023-11-09 20:42:24,184 | sd | DEBUG | sd_samplers | Sampler: sampler="KDPM2" config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'steps_offset': 1}
2023-11-09 20:42:24,193 | sd | DEBUG | processing_diffusers | Steps: type=base input=30 output=30 refiner=True
2023-11-09 20:42:24,268 | sd | DEBUG | processing_diffusers | Diffuser pipeline: StableDiffusionPipeline task=DiffusersTaskType.TEXT_2_IMAGE set={'prompt_embeds': torch.Size([1, 77, 768]), 'negative_prompt_embeds': torch.Size([1, 77, 768]), 'guidance_scale': 7, 'generator': device(type='cpu'), 'output_type': 'latent', 'num_inference_steps': 30, 'eta': 1.0, 'guidance_rescale': 0.7, 'height': 768, 'width': 768, 'parser': 'Full parser'}
2023-11-09 20:42:27,768 | sd | DEBUG | processing_diffusers | Moving to CPU: model=base
2023-11-09 20:42:28,101 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=Heun type=all config={'num_train_timesteps': 1000, 'beta_start': 0.0001, 'beta_end': 0.02, 'beta_schedule': 'linear', 'prediction_type': 'epsilon'}
2023-11-09 20:42:28,104 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=Heun type=scheduler config={'num_train_timesteps': 1000, 'beta_start': 0.0001, 'beta_end': 0.02, 'beta_schedule': 'linear', 'prediction_type': 'epsilon', 'use_karras_sigmas': False}
2023-11-09 20:42:28,106 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=Heun type=model config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'use_karras_sigmas': True}
2023-11-09 20:42:28,109 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=Heun type=user config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'use_karras_sigmas': True}
2023-11-09 20:42:28,137 | sd | DEBUG | sd_samplers | Sampler: sampler="Heun" config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'use_karras_sigmas': True}
2023-11-09 20:42:28,377 | sd | DEBUG | processing_diffusers | Steps: type=refiner input=10 output=5.0 start=0.5 denoise=0.5
2023-11-09 20:42:28,469 | sd | ERROR | processing_diffusers | Prompt parser encode: "LayerNormKernelImpl" not implemented for 'Half'
2023-11-09 20:42:28,470 | sd | DEBUG | processing_diffusers | Diffuser pipeline: StableDiffusionImg2ImgPipeline task=DiffusersTaskType.IMAGE_2_IMAGE set={'prompt': 1, 'negative_prompt': 1, 'guidance_scale': 6, 'generator': device(type='cpu'), 'output_type': 'latent', 'num_inference_steps': 5, 'eta': 1.0, 'image': <class 'torch.Tensor'>, 'parser': 'Fixed attention'}
2023-11-09 20:42:28,885 | sd | DEBUG | processing_diffusers | Moving to CPU: model=UNet
2023-11-09 20:42:29,089 | sd | DEBUG | processing_diffusers | VAE decode: name=vae-ft-mse-840000-ema-pruned.safetensors dtype=torch.float16 upcast=False images=1 latents=torch.Size([1, 4, 96, 96]) time=0.203
2023-11-09 20:42:29,092 | sd | DEBUG | processing_diffusers | Moving to CPU: model=refiner
2023-11-09 20:42:29,577 | sd | DEBUG | images | Saving: image="D:\StableDiffusion\outputs\text\2023-11-09\00069-20231109204229-8445c52d-b90a-401e-9e54-3b6e8a5ef0da.png" type=PNG size=768x768
2023-11-09 20:42:29,706 | sd | INFO | processing | Processed: images=1 time=5.74 its=5.23 memory={'ram': {'used': 6.58, 'total': 127.77}, 'gpu': {'used': 4.63, 'total': 23.99}, 'retries': 0, 'oom': 0}
2023-11-09 20:42:40,129 | sd | DEBUG | txt2img | txt2img: id_task=task(vb1fhx7rybsk0co)|prompt=((masterpiece)),((highres)),((best quality)),((highly detailed)),((extremely detailed CG unity 8k wallpaper)), detailed face, illustration,Sharon,solo,looking at viewer,crossed arms,expressionless,closed mouth,outdoors,dusk,street,from side|negative_prompt=(human, skin:1.3), worst quality, low res, bad anatomy, deformed, (extra limbs, extra legs, bad hands), extra ears, (easynegative, by bad-artist-anime, bad-hands-5, bad_prompt_version2)|prompt_styles=[]|steps=30|sampler_index=7|latent_index=14|full_quality=True|restore_faces=False|tiling=False|n_iter=1|batch_size=1|cfg_scale=7|clip_skip=2|seed=43534534.0|subseed=-1.0|subseed_strength=0|seed_resize_from_h=0|seed_resize_from_w=0||height=768|width=768|enable_hr=True|denoising_strength=0.5|hr_scale=2|hr_upscaler=None|hr_force=True|hr_second_pass_steps=30|hr_resize_x=0|hr_resize_y=0|image_cfg_scale=6|diffusers_guidance_rescale=0.7|refiner_steps=20|refiner_start=0.5|refiner_prompt=|refiner_negative=|override_settings_texts=[]
2023-11-09 20:42:40,144 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=all config={'num_train_timesteps': 1000, 'beta_start': 0.0001, 'beta_end': 0.02, 'beta_schedule': 'linear', 'prediction_type': 'epsilon'}
2023-11-09 20:42:40,146 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=scheduler config={'num_train_timesteps': 1000, 'beta_start': 0.0001, 'beta_end': 0.02, 'beta_schedule': 'linear', 'prediction_type': 'epsilon', 'steps_offset': 0}
2023-11-09 20:42:40,149 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=model config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'steps_offset': 1}
2023-11-09 20:42:40,153 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=user config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'steps_offset': 1}
2023-11-09 20:42:40,363 | sd | DEBUG | sd_samplers | Sampler: sampler="KDPM2" config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'steps_offset': 1}
2023-11-09 20:42:40,646 | sd | DEBUG | processing_diffusers | Steps: type=base input=30 output=30 refiner=True
2023-11-09 20:42:40,723 | sd | DEBUG | processing_diffusers | Diffuser pipeline: StableDiffusionPipeline task=DiffusersTaskType.TEXT_2_IMAGE set={'prompt_embeds': torch.Size([1, 77, 768]), 'negative_prompt_embeds': torch.Size([1, 77, 768]), 'guidance_scale': 7, 'generator': device(type='cpu'), 'output_type': 'latent', 'num_inference_steps': 30, 'eta': 1.0, 'guidance_rescale': 0.7, 'height': 768, 'width': 768, 'parser': 'Full parser'}
2023-11-09 20:42:44,309 | sd | DEBUG | processing_diffusers | Moving to CPU: model=base
2023-11-09 20:42:44,750 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=Heun type=all config={'num_train_timesteps': 1000, 'beta_start': 0.0001, 'beta_end': 0.02, 'beta_schedule': 'linear', 'prediction_type': 'epsilon'}
2023-11-09 20:42:44,753 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=Heun type=scheduler config={'num_train_timesteps': 1000, 'beta_start': 0.0001, 'beta_end': 0.02, 'beta_schedule': 'linear', 'prediction_type': 'epsilon', 'use_karras_sigmas': False}
2023-11-09 20:42:44,756 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=Heun type=model config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'use_karras_sigmas': True}
2023-11-09 20:42:44,759 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=Heun type=user config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'use_karras_sigmas': True}
2023-11-09 20:42:44,793 | sd | DEBUG | sd_samplers | Sampler: sampler="Heun" config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'use_karras_sigmas': True}
2023-11-09 20:42:45,141 | sd | DEBUG | processing_diffusers | Steps: type=refiner input=20 output=10.0 start=0.5 denoise=0.5
2023-11-09 20:42:45,234 | sd | ERROR | processing_diffusers | Prompt parser encode: "LayerNormKernelImpl" not implemented for 'Half'
2023-11-09 20:42:45,235 | sd | DEBUG | processing_diffusers | Diffuser pipeline: StableDiffusionImg2ImgPipeline task=DiffusersTaskType.IMAGE_2_IMAGE set={'prompt': 1, 'negative_prompt': 1, 'guidance_scale': 6, 'generator': device(type='cpu'), 'output_type': 'latent', 'num_inference_steps': 10, 'eta': 1.0, 'image': <class 'torch.Tensor'>, 'parser': 'Fixed attention'}
2023-11-09 20:42:46,134 | sd | DEBUG | processing_diffusers | Moving to CPU: model=UNet
2023-11-09 20:42:46,374 | sd | DEBUG | processing_diffusers | VAE decode: name=vae-ft-mse-840000-ema-pruned.safetensors dtype=torch.float16 upcast=False images=1 latents=torch.Size([1, 4, 96, 96]) time=0.239
2023-11-09 20:42:46,378 | sd | DEBUG | processing_diffusers | Moving to CPU: model=refiner
2023-11-09 20:42:46,911 | sd | DEBUG | images | Saving: image="D:\StableDiffusion\outputs\text\2023-11-09\00070-20231109204246-426fbce6-ba59-4e65-afbe-e5c99c131462.png" type=PNG size=768x768
2023-11-09 20:42:47,094 | sd | INFO | processing | Processed: images=1 time=6.95 its=4.32 memory={'ram': {'used': 6.64, 'total': 127.77}, 'gpu': {'used': 4.63, 'total': 23.99}, 'retries': 0, 'oom': 0}
2023-11-09 20:42:50,018 | sd | DEBUG | txt2img | txt2img: id_task=task(wv2li3g7nb17io9)|prompt=((masterpiece)),((highres)),((best quality)),((highly detailed)),((extremely detailed CG unity 8k wallpaper)), detailed face, illustration,Sharon,solo,looking at viewer,crossed arms,expressionless,closed mouth,outdoors,dusk,street,from side|negative_prompt=(human, skin:1.3), worst quality, low res, bad anatomy, deformed, (extra limbs, extra legs, bad hands), extra ears, (easynegative, by bad-artist-anime, bad-hands-5, bad_prompt_version2)|prompt_styles=[]|steps=30|sampler_index=7|latent_index=14|full_quality=True|restore_faces=False|tiling=False|n_iter=1|batch_size=1|cfg_scale=7|clip_skip=2|seed=43534534.0|subseed=-1.0|subseed_strength=0|seed_resize_from_h=0|seed_resize_from_w=0||height=768|width=768|enable_hr=True|denoising_strength=0.5|hr_scale=2|hr_upscaler=None|hr_force=True|hr_second_pass_steps=30|hr_resize_x=0|hr_resize_y=0|image_cfg_scale=6|diffusers_guidance_rescale=0.7|refiner_steps=30|refiner_start=0.5|refiner_prompt=|refiner_negative=|override_settings_texts=[]
2023-11-09 20:42:50,033 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=all config={'num_train_timesteps': 1000, 'beta_start': 0.0001, 'beta_end': 0.02, 'beta_schedule': 'linear', 'prediction_type': 'epsilon'}
2023-11-09 20:42:50,035 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=scheduler config={'num_train_timesteps': 1000, 'beta_start': 0.0001, 'beta_end': 0.02, 'beta_schedule': 'linear', 'prediction_type': 'epsilon', 'steps_offset': 0}
2023-11-09 20:42:50,037 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=model config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'steps_offset': 1}
2023-11-09 20:42:50,040 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=KDPM2 type=user config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'steps_offset': 1}
2023-11-09 20:42:50,243 | sd | DEBUG | sd_samplers | Sampler: sampler="KDPM2" config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'steps_offset': 1}
2023-11-09 20:42:50,496 | sd | DEBUG | processing_diffusers | Steps: type=base input=30 output=30 refiner=True
2023-11-09 20:42:50,566 | sd | DEBUG | processing_diffusers | Diffuser pipeline: StableDiffusionPipeline task=DiffusersTaskType.TEXT_2_IMAGE set={'prompt_embeds': torch.Size([1, 77, 768]), 'negative_prompt_embeds': torch.Size([1, 77, 768]), 'guidance_scale': 7, 'generator': device(type='cpu'), 'output_type': 'latent', 'num_inference_steps': 30, 'eta': 1.0, 'guidance_rescale': 0.7, 'height': 768, 'width': 768, 'parser': 'Full parser'}
2023-11-09 20:42:54,130 | sd | DEBUG | processing_diffusers | Moving to CPU: model=base
2023-11-09 20:42:54,438 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=Heun type=all config={'num_train_timesteps': 1000, 'beta_start': 0.0001, 'beta_end': 0.02, 'beta_schedule': 'linear', 'prediction_type': 'epsilon'}
2023-11-09 20:42:54,440 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=Heun type=scheduler config={'num_train_timesteps': 1000, 'beta_start': 0.0001, 'beta_end': 0.02, 'beta_schedule': 'linear', 'prediction_type': 'epsilon', 'use_karras_sigmas': False}
2023-11-09 20:42:54,442 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=Heun type=model config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'use_karras_sigmas': True}
2023-11-09 20:42:54,444 | sd | DEBUG | sd_samplers_diffusers | Sampler: name=Heun type=user config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'use_karras_sigmas': True}
2023-11-09 20:42:54,472 | sd | DEBUG | sd_samplers | Sampler: sampler="Heun" config={'num_train_timesteps': 1000, 'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'prediction_type': 'epsilon', 'use_karras_sigmas': True}
2023-11-09 20:42:54,725 | sd | DEBUG | processing_diffusers | Steps: type=refiner input=30 output=15.0 start=0.5 denoise=0.5
2023-11-09 20:42:54,806 | sd | ERROR | processing_diffusers | Prompt parser encode: "LayerNormKernelImpl" not implemented for 'Half'
2023-11-09 20:42:54,808 | sd | DEBUG | processing_diffusers | Diffuser pipeline: StableDiffusionImg2ImgPipeline task=DiffusersTaskType.IMAGE_2_IMAGE set={'prompt': 1, 'negative_prompt': 1, 'guidance_scale': 6, 'generator': device(type='cpu'), 'output_type': 'latent', 'num_inference_steps': 15, 'eta': 1.0, 'image': <class 'torch.Tensor'>, 'parser': 'Fixed attention'}
2023-11-09 20:42:56,145 | sd | DEBUG | processing_diffusers | Moving to CPU: model=UNet
2023-11-09 20:42:56,412 | sd | DEBUG | processing_diffusers | VAE decode: name=vae-ft-mse-840000-ema-pruned.safetensors dtype=torch.float16 upcast=False images=1 latents=torch.Size([1, 4, 96, 96]) time=0.266
2023-11-09 20:42:56,419 | sd | DEBUG | processing_diffusers | Moving to CPU: model=refiner
2023-11-09 20:42:57,026 | sd | DEBUG | images | Saving: image="D:\StableDiffusion\outputs\text\2023-11-09\00071-20231109204257-9e2b6304-db13-4a01-ba02-a1116e0e369b.png" type=PNG size=768x768
2023-11-09 20:42:57,221 | sd | INFO | processing | Processed: images=1 time=7.19 its=4.17 memory={'ram': {'used': 6.47, 'total': 127.77}, 'gpu': {'used': 4.63, 'total': 23.99}, 'retries': 0, 'oom': 0}
Issue Description
This one is relatively straightforward. The refiner steps for the SDXL pipeline are stuck at 99 regardless of the settings chosen in the web UI.
2023-11-08 22:51:10,748 | sd | DEBUG | processing_diffusers | Diffuser pipeline: StableDiffusionXLImg2ImgPipeline task=DiffusersTaskType.IMAGE_2_IMAGE set={'prompt_embeds': torch.Size([1, 77, 1280]), 'pooled_prompt_embeds': torch.Size([1, 1280]), 'negative_prompt_embeds': torch.Size([1, 77, 1280]), 'negative_pooled_prompt_embeds': torch.Size([1, 1280]), 'guidance_scale': 6, 'generator': device(type='cpu'), 'output_type': 'latent', 'num_inference_steps': 99, 'eta': 1.0, 'guidance_rescale': 0.7, 'denoising_start': 0.8, 'denoising_end': 1, 'image': <class 'torch.Tensor'>, 'parser': 'Full parser'}
Version Platform Description
Windows 10, Nvidia RTX 4090, Diffusers backend, Firefox, master branch. e3cd9e47.
Relevant log output
Backend
Diffusers
Model
SD-XL
Acknowledgements