Error Running Stable Diffusion from the command line in Windows

inkblot 0
Viewed 1281

I installed Stable Diffusion v1.4 by following the instructions described in https://www.howtogeek.com/830179/how-to-run-stable-diffusion-on-your-pc-to-generate-ai-images/#autotoc_anchor_2

My machine heavily exceeds the min reqs to run Stable Diffusion: Windows 11 Pro 11th Gen Intel i7 @ 2.30GHz Latest NVIDIA GeForce GPU 16GB Memory 1TB SSD

Yet, I get an error when trying to run the test prompt python scripts/txt2img.py --prompt "a close-up portrait of a cat by pablo picasso, vivid, abstract art, colorful, vibrant" --plms --n_iter 5 --n_samples 1

RuntimeError: CUDA out of memory. Tried to allocate 1024.00 MiB (GPU 0; 8.00 GiB total capacity; 6.13 GiB already allocated; 0 bytes free; 6.73 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

Reading a post by Marco Ramos it seems like it relates to the number of workers in PyTorch Strange Cuda out of Memory behavior in Pytorch

How do I change the number of workers while running Stable Diffusion? And why is it throwing this error if my machine still has lots of memory? Has anyone encountered this same issue while running Stable Diffusion?

command-line-interface windows pytorch python
2 Answers

I had the same issue, it's because you're using a non-optimized version of Stable-Diffusion. You have to download basujindal's branch of it, which allows it use much less ram by sacrificing the precision, this is the branch - https://github.com/basujindal/stable-diffusion Everything else in that guide stays the same just clone from this version. It allow you to even push past 512x512 default resolution, you can use 756x512 to get rectangular images for example (but the results may vary since it was trained on a 512 square set).

the new prompt becomes python optimizedSD/optimized_txt2img.py --prompt "blue orange" --H 756 --W 512

Also another note: as of a few days ago an even faster and more optimized version was released by neonsecret (https://github.com/basujindal/stable-diffusion), however I'm having issues installing it, so can't really recommend it but you can try it as well and see if it works for you.

OortCloud210

In addition to the optimized version by basujindal, the additional tags following the prompt allows the model to run properly on a machine with NVIDIA or AMD 8+GB GPU. So the new prompt would look like this >> python optimizedSD/optimized_txt2img.py --prompt "a close-up portrait of a cat by pablo picasso, vivid, abstract art, colorful, vibrant" --H 512 --W 512 --seed 27 --n_iter 2 --n_samples 10 --ddim_steps 50

inkblot0
Related