YanWenKun/ComfyUI-Docker v6.0

on GitHub

It’s been a long time since the last release, and many major events have taken place in the AIGC field. This project itself has also undergone significant updates — it’s time to mark a new milestone.

Added Image Tags

• cpu
• cu126-slim
• cu128-slim
• cu130-slim
• cu126-megapak
• cu128-megapak
• xpu
• xpu-cn

Retired Image Tags

• cu118-slim
• cu118-megapak
  • Retired with PyTorch 2.7.1
• cu121
• cu121-megapak
  • Retired with PyTorch 2.5.1
• cu124-cn
• cu124-slim
• cu124-megapak
  • Retired with PyTorch 2.6.0
• cu128-test
• cu129-slim
• cu129-test
• cu129-megapak
  • Retired with PyTorch 2.8.0
• comfy3d-pt22
• comfy3d-pt23
• comfy3d-pt25
• direct3ds2

Planned Image Tags

• cu130-megapak
  • Planned to use Python 3.14t (free-threading, no GIL). Waiting for the ecosystem.
• cu128-cn
• rocm-cn

Features for New Image Tags

• Speed up startup time. First starts no longer download anything from the Internet.

  • Now a ComfyUI bundle is preloaded in the image for copy on-demand.
  • No longer download "out-of-box" SD models. It's not necessary now.

• Mount model files and user files separately.

• Speed up image pulling by breaking down the steps in Dockerfile to have more smaller image layers.

• All changes are backward-compatible.

Minor Changes

• Dropped mediapipe in all images. Ref: google-ai-edge/mediapipe#5810

• Full Changelog: v5.0...v6.0

Thanks

• Thanks @nhtua for #67 (critical ROCm env vars)

• Thanks @seasuu for #101 (fix libgl for Ubuntu, now replaced by Fedora)

• Thanks @DarthGibblet for #110 (NVIDIA cusparselt newly introduced in PyTorch)

• Thanks @SergeyFilippov for #114 (update rocm to 6.4.2)

• Thanks for the information in issues #79 #81 #85 #87 #90 #107 #124 #126 #128 #130 #131