stable-diffusion
stable-diffusion copied to clipboard
Is this fork still relevant with Stable Diffusion V2?
Is the new release more optimized than the original?
If it is now, is there any way to use these changes in the new release?
Stable diffusion 2 is a totally different repository So the question is if there's going to be a new fork with similar improvements.
Stable diffusion 2 is a totally different repository So the question is if there's going to be a new fork with similar improvements.
I have come to learn that at least some of these optimizations (and more) are in the WebUI stable diffusion fork: https://github.com/AUTOMATIC1111/stable-diffusion-webui/
There is also a newer fork of this very repo, but I cant get it to work and it has GH issues disabled.
That updated repo is here: https://github.com/consciencia/stable-diffusion
There are so many SD forks that they are almost impossible to explore.
I believe this fork is still relevant with Stable Diffusion V2 because the V2 model is even larger and requires more memory to load in whole. This fork works well with V1.4 and V1.5, as I currently have it running at https://aitransformer.net/SuperStylizer, using the V1.5 model and the server only has 8G VRAM. I inspected V2 code at https://github.com/Stability-AI/stablediffusion, the changes seem not too big, so I managed to update the code using the same way of splitting the model to run on GPU one part at a time, the model loading is fine but unfortunately the process got killed every time when I tried to run a simple txt2img, probably due to running out of memory.
...... UNet: Running in eps-prediction mode Setting up MemoryEfficientCrossAttention. Query dim is 320, context_dim is None and using 5 heads. Setting up MemoryEfficientCrossAttention. Query dim is 640, context_dim is None and using 10 heads. ...... CondStage: Running in eps-prediction mode Killed
It'd be great if the author can make it working for V2. I can share the code changes I have so far if needed.