Streamlined interface for generating images with AI in Krita. Inpaint and outpaint with optional text prompt, no tweaking required. - GitHub - Acly/krita-ai-diffusion: Streamlined interface for gen…

  • wim@lemmy.sdf.org
    link
    fedilink
    arrow-up
    13
    ·
    1 year ago

    There have been some efforts to run pytorch and StableDiffusion on ROCm. Not sure if that could be combined with this.

    • wewbull@feddit.uk
      link
      fedilink
      English
      arrow-up
      5
      ·
      edit-2
      1 year ago

      It works today. Only problem I have is the memory management is pretty poor, and it’s pretty easy to run out of vram.

      Rx7600 8GB + 5900X Rocm 5.7.1 Pytorch 2.1

      • wim@lemmy.sdf.org
        link
        fedilink
        arrow-up
        1
        ·
        edit-2
        1 year ago

        Interesting! Got any links that explain how to set it up?

        I just got a laptop with an RX 6700M 10GB ans am eager to try it :)

        • wewbull@feddit.uk
          link
          fedilink
          English
          arrow-up
          4
          ·
          1 year ago

          Not really. I’ve had to do quite a bit of experimentation.

          My setup that I’ve settled on:

          • Rocm system libraries from Arch Linux
          • PyTorch nightly for Rocm pip installed into a venv (see instructions on pytorch homepage)
          • Set HSA_OVERRIDE_GFX_VERSION to 11.0.0. This is just for the RX7600 and it tells it to use the RX7900 code as the pytorch version hasn’t been compiled with 7600 support.
          • Start software.
    • ByteSeb@beehaw.org
      link
      fedilink
      arrow-up
      2
      ·
      11 months ago

      Hmm, that’s weird. I was able to run Stable Diffusion locally with Linux + RX6600.

      Probably because I used Easy Diffusion. At first, I couldn’t get the GPU acceleration to work, and I was constantly running out of RAM (Not using VRAM), so my system always froze and crashed.

      Turns out it was a ROCM bug, that I don’t know if it’s fixed by now, but I remember “fixing it” by setting an environment variable to a previous version.

      Then, it all worked really good. Took between 30 seconds to 2 minutes to make an image.