AMD ROCm GPU Computing Stack

Packages that are neither part of ROCm nor on (some of) the
build targets but required for the ROCm build

This project provides packages related to Machine Learning.

ComfyUI is a node-graph engine for local AI content creation. Workflows
are assembled from nodes that load diffusion models, encoders, LoRAs and
other assets, and can produce images, video, audio and 3D output. A local
HTTP API and a browser UI are included.

The packaged tree is read-only. Models, custom nodes, input, output, temp
and user state live under the per-user --base-directory (default:
$XDG_DATA_HOME/comfyui or ~/.local/share/comfyui); the custom nodes shipped
with ComfyUI itself are loaded from the package tree as well.
ComfyUI-Manager is not enabled.

A Python library that provides an open()-like interface for reading
local and remote files (Google Cloud Storage and Azure Blob Storage),
plus os.path- and shutil-like helpers that also accept GCS (gs://)
and Azure (az://) paths. Inspired by TensorFlow's gfile, but not an
exact clone of that API.

AIMDO is a PyTorch VRAM allocator used by ComfyUI to offload model weights under memory pressure. This build ships the pure-Python fallback. The native CUDA/HIP allocator is not compiled because it requires CUDA 12.8+ and a CUDA-enabled PyTorch, which Factory does not provide.

Distro implementation of the comfy-angle 0.1.0 Python API used by
ComfyUI's GLSL Shader node. The module locates the system's libEGL
and libGLESv2 (libglvnd dispatch plus Mesa) at runtime. It does not
ship Chromium ANGLE or Electron-extracted binaries.

Comfy Kitchen is a kernel library for diffusion-model inference used by ComfyUI. This build ships the pure-Python eager (CPU) backend. Native CUDA and HIP extensions are not compiled: Factory python-torch is CPU-only and has no CUDA flavor.

Embedded documentation for ComfyUI nodes.

Official ComfyUI frontend static assets shipped as a Python package.

Meta package that re-exports the ComfyUI workflow template helpers
and pulls in the core, JSON and media asset bundles at the versions
this release is pinned to.

Core helpers for ComfyUI workflow templates: manifest loading, asset
path resolution and metadata used by the media bundles and the
comfyui-workflow-templates meta package.

Workflow template JSON definitions used by the ComfyUI template
picker.

Media bundle containing API-driven workflow assets for ComfyUI.

Media assets bundle 01 for ComfyUI workflow templates.

Media bundle containing image workflow assets for ComfyUI.

Media bundle containing audio, 3D and miscellaneous workflow assets for ComfyUI.

Media bundle containing video workflow assets for ComfyUI.

EasyDict allows accessing dict values as attributes (works
recursively). It provides Javascript-like properties dot notation
for Python dicts.

Python library for reading and writing ML models in the GGUF (GGML
Universal File) format used by GGML-based inference engines such as
llama.cpp. It also ships command-line tools to dump metadata, edit
metadata values, copy a GGUF file with new metadata, and convert
endianness.

Client library to download and publish models, datasets and other repos on the huggingface.co hub

Kornia is a differentiable computer vision library built on PyTorch.

Python bindings for kornia-rs, a low-level computer vision library written in Rust.