Run on Modal
The ComfyUI workers also run on Modal, another serverless GPU service billed by the second, from modal_app.py in each repo.
It builds the same Dockerfile and answers the same requests as on RunPod (/run, /status/<id>), so a client that calls RunPod endpoints needs only the other address and key.
The Models page shows which workers have it.
On your computer, with Python 3.10 or newer, in a copy of your repo (git clone it, with your Dockerfile lines filled in):
pip install modalmodal setup # opens a page to sign in to Modalmodal secret create cobaltpear-key CP_KEY=<a long password you make up>modal deploy modal_app.py # builds, then prints the worker's address- The address ending in
-api….modal.runis your endpoint; sendAuthorization: Bearer <that password>with each request. - GPU: the
GPUline at the top ofmodal_app.py— set to the smallest one that fits the model. A worker turns off 60 s after its last picture. - Modal needs a payment method on your account to use GPUs. Its prices: modal.com/pricing.
- Updates: run
modal deploy modal_app.pyagain.