stable-diffusion-webui
stable-diffusion-webui copied to clipboard
AUTOMATIC1111
[Bug]: Suddenly DPM img2img is all blurry
Is there an existing issue for this?
- [X] I have searched the existing issues and checked the recent builds/commits
What happened?
I've been using img2img like normally, mostly with ControlNet, everything was fine, but suddenly today every result is blurry when I use DPM on macOS.
When I switch to Euler A, I get the sharp result.
Steps to reproduce the problem
When using Euler A, compare to DDM Karras, etc.
What should have happened?
It should not be blurry.
Commit where the problem happens
When doing img2img
What platforms do you use to access the UI ?
MacOS
What browsers do you use to access the UI ?
Apple Safari
Command Line Arguments
No
List of extensions
ControlNet
Console logs
ython 3.10.10 (main, Feb 8 2023, 05:34:50) [Clang 14.0.0 (clang-1400.0.29.202)]
Commit hash: 0cc0ee1bcb4c24a8c9715f66cede06601bfc00c8
Installing requirements for Web UI
Installing requirement for sd-webui-controlnet
Installing sd-dynamic-prompts requirements.txt
Launching Web UI with arguments: --no-half --upcast-sampling --no-half-vae --use-cpu interrogate
Warning: caught exception 'Torch not compiled with CUDA enabled', memory monitor disabled
No module 'xformers'. Proceeding without it.
==============================================================================
You are running torch 1.12.1.
The program is tested to work with torch 1.13.1.
To reinstall the desired version, run with commandline flag --reinstall-torch.
Beware that this will cause a lot of large files to be downloaded, as well as
there are reports of issues with training tab on the latest version.
Use --skip-version-check commandline argument to disable this check.
==============================================================================
Loading weights [fc2511737a] from /Users/jimmygunawan/stable-diffusion-webui/models/Stable-diffusion/chilloutmix_NiPrunedFp32Fix.safetensors
Creating model from config: /Users/jimmygunawan/stable-diffusion-webui/configs/v1-inference.yaml
LatentDiffusion: Running in eps-prediction mode
DiffusionWrapper has 859.52 M params.
Applying cross attention optimization (InvokeAI).
Textual inversion embeddings loaded(1): ulzzang-6500
Model loaded in 6.6s (create model: 0.5s, apply weights to model: 4.7s, move model to device: 1.2s).
Running on local URL: http://127.0.0.1:7860
To create a public link, set `share=True` in `launch()`.
Loading preprocessor: none, model: control_sd15_openpose [fef5e48e]
Loaded state_dict from [/Users/jimmygunawan/stable-diffusion-webui/extensions/sd-webui-controlnet/models/control_sd15_openpose.pth]
ControlNet model control_sd15_openpose [fef5e48e] loaded.
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:34<00:00, 1.73s/it]
Total progress: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:28<00:00, 1.40s/it]
Will process 89 images, creating 1 new images for each.
INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will create 1 images in a total of 1 batches.
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:48<00:00, 2.43s/it]
INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will create 1 images in a total of 1 batches. | 20/1780 [00:29<42:52, 1.46s/it]
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:47<00:00, 2.38s/it]
INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will create 1 images in a total of 1 batches. | 40/1780 [01:30<47:59, 1.65s/it]
5%|███████ | 1/20 [00:13<04:12, 13.29s/it]
Total progress: 2%|██▊ | 41/1780 [02:01<1:25:53, 2.96s/it]
Loading weights [c35782bad8] from /Users/jimmygunawan/stable-diffusion-webui/models/Stable-diffusion/realisticVisionV13_v13.safetensors
Applying cross attention optimization (InvokeAI).
Weights loaded in 7.2s (load weights from disk: 0.4s, apply weights to model: 4.8s, move model to device: 2.0s).
INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will create 1 images in a total of 1 batches.
restoring last networks
Loading preprocessor: openpose, model: control_sd15_openpose [fef5e48e]
Loaded state_dict from [/Users/jimmygunawan/stable-diffusion-webui/extensions/sd-webui-controlnet/models/control_sd15_openpose.pth]
ControlNet model control_sd15_openpose [fef5e48e] loaded.
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:40<00:00, 2.03s/it]
Total progress: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:30<00:00, 1.51s/it]
INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will create 1 images in a total of 1 batches.
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:34<00:00, 1.74s/it]
Total progress: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:31<00:00, 1.56s/it]
INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will create 1 images in a total of 1 batches.
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:37<00:00, 1.88s/it]
Total progress: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:33<00:00, 1.69s/it]
Will process 89 images, creating 1 new images for each.
INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will create 1 images in a total of 1 batches.
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:37<00:00, 1.86s/it]
INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will create 1 images in a total of 1 batches. | 20/1780 [00:31<48:28, 1.65s/it]
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:34<00:00, 1.74s/it]
INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will create 1 images in a total of 1 batches. | 40/1780 [01:11<42:44, 1.47s/it]
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [00:38<00:00, 1.92s/it]
INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will creat
Additional information
When using DPM all my result is blurry like it's censored or something:
This may not be the cause of your problem, but I noticed that your example image has 'Noise Multiplier: 0' in its metadata, which doesn't seem normal.
Did your settings get messed up somehow? Try with a different Noise multiplier for img2img in the global settings to see if the problem remains.
When I import the image with the PNG Info tab, it even tries to give a global override for 'Noise multiplier: 0.0', which doesn't actually work and is ignored (my default of 0.75 is used), and the images generated seem sharp (I only did a little testing, though).
Furthermore, I cannot even manually set 0 for that value in the settings. For me, 0.5 is the minimum it lets me choose. I did a very quick test and using 0.5 does seem blurrier for DPM samplers compared to Euler a, so maybe 0.0 would explain what you see?
I guess it's also possible that it was just an error with how the metadata got written to the file, and the multiplier wasn't actually 0 or related to your issue...
I am having exactly the same problem, since last night. I am not using Control Net though, and I am getting blurry Img2Img results using all samplers, and any checkpoint. I don't know what might have happened, I don't think I touched any settings, and my noise multipler is 0.54... I can't change it using the slider at the top of the screen, and I can't find it in Global Settings. If anyone has any idea why this could have happened, or what info would be useful to provide, I'd be very helpful because currently my entire workflow is rendered useless.
Weird.. what is the metadata of an example image that generated with the problem? Maybe there is some sort of corruption of the settings, but that's pure speculation.
You can just open the file with something like Notepad++ and take a screenshot.
Thanks @EfourC for the suggestion. I did it, and for some reason the actual model does not appear in my PNG header like it does in yours. Maybe that could be the problem, it makes sense, but I don't know why the model wouldn't be there. I will try again tonight.
Thanks.. hmm your metadata looks ok to me so I'm not sure what could be causing the blurriness in your images.
As for the model name, I believe i it's just an option you can enable. However, it being enabled or disabled wouldn't affect image generation. I hadn't thought about it when I took my screenshot, but it also shows testing out code changes that add separate highres prompts, so just ignore those too.
Have you found a solution to your problem?
I would suggest checking if your Noise multiplier for img2img is less than 1
Find or set this function in settings > user interface > quicksettings list > write this command - initial_noise_multiplier
I'm having similar issues except it happens with all samplers and Euler A doesn't give sharp results.
Latent hires was making the blur for me, i changed it to ESRGAN 4x (web ui with python 2.0.1)
It was the noise multiplier causing this for me. Had to turn it up above 1
