Jetson-Containers for Stable-Diffusion3 / from diffusers import StableDiffusion3Pipeline

[kyle-redyeti]

Search before asking

• I have searched the jetson-containers issues and found no similar feature requests.

Question

I have been attempting to find the "magic sauce" for utilizing PyTorch with CUDA and StableDiffusion3Pipeline in a container on my Orin AGX. I was able to put together a super simple set of pip installs to get StableDiffusion3Pipeline working with a version of PyTorch that is not CUDA enabled (what ever version it will install with pip install -U torch diffusers huggingface_hub transformers accelerate.

In my adventures to use torch with CUDA so image generation takes less that 20 hours I seem to get caught in odd "dependency hell" where diffusion (I am pretty sure it is diffusion) always wants a different version of torch???

Has anyone else made this work and could share the simple Dockerfile? My guess is that I need a new version torch than my docker container was creating...

Also, it looks like diffusers has been deprecated in the repo??? Is that because it is going to be upgraded to a newer version or is it just not going to be (can't be )supported???

Thanks!

Additional

No response