iohann.s.titov@gmail.com Ivan S. Titov Build the torch backend (sci-ml/caffe2) with CUDA/NVIDIA acceleration; also selects the CUDA wheels of dev-python/comfy-aimdo-bin and dev-python/comfy-kitchen-bin (otherwise their non-CUDA wheels) Build the torch backend (sci-ml/caffe2) with AMD ROCm/HIP acceleration; installs the non-CUDA wheels of dev-python/comfy-aimdo-bin and dev-python/comfy-kitchen-bin and pulls dev-python/triton-bin, whose Triton backend runs comfy_kitchen.apply_rope on AMD. Mutually exclusive with cuda Pull dev-python/triton-bin for torch.compile / inductor custom nodes and comfy_kitchen's Triton backend Install the bundled example-workflow template gallery (dev-python/comfyui-workflow-templates; ~330 MiB of preview assets) Pull optional model helpers — dev-python/spandrel (upscaler architectures) and dev-python/kornia (computer-vision nodes) Pull dev-python/pyopengl + dev-python/glfw for the 3D-preview nodes Pull sci-ml/torchaudio for the audio nodes (audio encoders, audio VAE). torchaudio must match your installed sci-ml/pytorch — it currently lags the 2.12 stack, so this may not resolve yet comfy-org/comfyui