qunash / stable-diffusion-2-gui

Lightweight Stable Diffusion v 2.1 web UI: txt2img, img2img, depth2img, inpaint and upscale4x.

Stable Diffusion v 2.1 web UI

Gradio app for Stable Diffusion 2 by Stability AI. It uses Hugging Face Diffusers🧨 implementation.

Currently supported pipelines are text-to-image, image-to-image, inpainting, upscaling and depth-to-image.

Colab by anzorq. If you like it, please consider supporting me:

---

Credits

• Stable Diffusion - https://github.com/CompVis/stable-diffusion
• HuggingFace Diffusers🧨 - https://github.com/huggingface/diffusers
• DPM-Solver - https://github.com/LuChengTHU/dpm-solver

stable-diffusion-2-gui's Issues

timeout on collab?

why gradio keeping time out?

Is it possible to lift off the 4x upscaler restriction only for 512x512 images

Your 4x upscaler did a very good job at upscaling the detail of the eyes, that I cannot recreate that with stable-diffusion-webui and Stable Diffusion V2.1. However, there are visible tiles in the upscaled photo, is it possible to lift off that 512 x512 restriction?

Connection errored out

Randomly losing the connection, without any info in the output cell. I have to restart the cell to fix the issue

can't run app on colab,got error.

when I click play button of run the app section,got an error:

OSError: [Errno 36] File name too long: '.main-div div{display:inline-flex;align-items:center;gap:.8rem;font-size:1.75rem}.main-div div h1{font-weight:900;margin-bottom:7px}.main-div p{margin-bottom:10px;font-size:94%}a{text-decoration:underline}.tabs{margin-top:0;margin-bottom:0}#gallery{min-height:20rem}\n'

how can we use another model?

thank you for your good work. i wonder how can i use anpther model (ex: epicrealism drom civitai)

Error when wanting to upscale

I kept getting this:
upscale() takes 7 positional arguments but 8 were given

I think 4x upscale should just be a button. Not drag and drop and needing prompting. But I am curious how it could actually work, if it can add details while upscaling.

Colab - Cuda OOM

After trying Depth-to-Image model a 4 or 5 times, I got this error.

CUDA out of memory. Tried to allocate 6.78 GiB (GPU 0; 14.76 GiB total capacity; 6.31 GiB already allocated; 5.71 GiB free; 7.80 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

I'm not sure if something wrong happens with the D2I model. Just letting you know!

Local Models

Where in the code should local models be specified?

Colab not working

Hey, as I run the colab cells I'm getting this error:

NameError Traceback (most recent call last) [<ipython-input-2-c92a6aca1dc2>](https://localhost:8080/#) in <module> 413 414 gallery = gr.Gallery(label="Generated images", show_label=False).style(grid=[2], height="auto") --> 415 state_info = gr.Textbox(label="State", show_label=False, max_lines=2, interactive=false).style(container=False) 416 error_output = gr.Markdown(visible=False) 417

Maximum seed length?

It seems like everything after the first 288 characters of the prompt is discarded. Is this correct? Is there a way to raise this limit? Thanks.

Upscaling yields broken images

Hi,

thanks for building this. I managed to use the upscaler on a 768x image the first time I used the plugin but after that I get broken images after upscaling, also for 512x. The image is displayed in preview but when I try to open / download it, it breaks halfway through loading.

• Do you think the reason might be related to limited RAM in colabs?
• Does the upscaler work for you?

Sampling methods

Hello, I wanted to ask which is the sampling method used for this tool, as I was unable to find any reference in the code

I was getting only black images.

Only black images for whatever prompt

This could be a potential solution

https://www.reddit.com/r/StableDiffusion/comments/zf21db/comment/iz9sh0i/?utm_source=reddit&utm_medium=web2x&context=3

Show the seed of the generated image

I have tried many notebooks and yours is the simplest, fastest and most stable of them all! The only function I miss is to see the seed of the image I just generated

Cannot run in AMD as there is no xformers

Is it possible to exclude the xformers check, this is the main reason AMD user needs to run SD in Colab instead of running in local

Incompatible torch

After about 2 weeks from last usage, I ran the code, but now I get an error that doesn't let to use SD.

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. torchaudio 2.0.2+cu118 requires torch==2.0.1, but you have torch 1.13.1 which is incompatible. torchdata 0.6.1 requires torch==2.0.1, but you have torch 1.13.1 which is incompatible. torchtext 0.15.2 requires torch==2.0.1, but you have torch 1.13.1 which is incompatible. torchvision 0.13.1+cu113 requires torch==1.12.1, but you have torch 1.13.1 which is incompatible.

When creating in 'Text to Image', it shows a few other errors and the waiting icon keeps infinitely long. Setting to img2img or other options, it doesn't rearrange the box (the field to upload an image file doesn't even show up).

Missing Licence

This is a really nice bit of code. Have you considered adding a licence (so others can build on it) or is to remain closed source?

img2img error

I cant get img2img to work. I am getting this error:

Error
Pipeline <class 'diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline'> expected {'unet', 'text_encoder', 'scheduler', 'feature_extractor', 'vae', 'safety_checker', 'tokenizer'}, but only {'unet', 'text_encoder', 'scheduler', 'vae', 'tokenizer'} were passed.

Crash when switching mode

The runtime crashes sometime when switching between 2 modes, and in my trials, everytime I switch from img2img to text2img