Run ComfyUI on your own GPU pool.
Queue workflows from one control plane. Run them on the AMD and NVIDIA machines you already have.
Save a workflow
Import a ComfyUI API graph and its Etai spec. Reuse a named version with new inputs.
Schedule on either GPU
Connect CUDA and ROCm workers. Pick a Worker for a run, or let Etai find an available GPU and prefer one with the weights cached.
Keep data local
Self-host the control plane with Docker Compose. Workers fetch the weights with Orme and return the outputs to your server.
Download Etai
Start your own control plane with Docker Compose.