C2C Farm Submit — Remote Render

Here's the honest pitch: C2C_Submit turns your ComfyUI graph into a client that hands a whole workflow to another ComfyUI instance - a LAN workstation, an AKS H100 pool, a RunPod pod - and waits for the result. You paste a workflow in ComfyUI API format into the node, pick a backend, and the pack's C2C Farm spooler does preflight validation, uploads your media, dispatches the job, and pulls the outputs back into a local results folder. It's the closest thing to a distributed render button the ecosystem has without standing up a commercial queue.

The catch, and you should hear it up front: this node is the front door of a subsystem that needs real configuration before it's useful. It's not an "install and go" node - it's the client half of a small render farm. If you already run multiple ComfyUI boxes, it's genuinely great. If you're a single-machine hobbyist, keep reading for the gotchas before you wire it in.

How it works

The submit path in nodes/render_farm.py is a well-structured pipeline. It looks up the backend from renderfarm/config/backends.json, validates the compute profile against what the backend accepts, checks your user's project permissions, parses the pasted prompt JSON, runs a preflight node check against the backend (so you fail fast instead of failing mid-render on the far side), uploads any referenced media through the courier, and spools a Job into the local queue manager. With wait_for_completion on, the node blocks - streaming progress events - until the job finishes, then returns the result paths. With it off, you get the job id immediately and track it in the C2C Farm dashboard or via C2C_JobHistory.

The inputs that matter

• prompt_json - the workflow in ComfyUI API format. Export with "Save (API Format)" and paste the JSON. It accepts a bare {node_id: {...}} map or a {'prompt': {...}} wrapper.
• backend - which registered backend (the list comes from backends.json).
• compute_profile - heavy_94gb (1 replica per 94GB GPU) or light_30gb (3 replicas per GPU).
• priority (5, 1–10) - higher dispatches first.
• wait_for_completion (true) - block-and-pull, or fire-and-track.
• timeout_minutes (120) - if it exceeds this, the node raises, but the job keeps running.

Outputs: job_id, result_paths, and info (JSON with status and results).

Installing it

Part of Code2Collapse/ComfyUI-CustomNodePacks. ComfyUI Manager → search "CustomNodePacks", or:

cd ComfyUI/custom_nodes
git clone https://github.com/Code2Collapse/ComfyUI-CustomNodePacks.git

Then the farm-specific setup: pip install requests websocket-client (plus a storage SDK only if you use cloud couriers), set the C2C_USER_NAME environment variable to a user defined in renderfarm/config/users.json, and enable at least one backend in renderfarm/config/backends.json.

Gotchas

Three things bite everyone. First: no C2C_USER_NAME set means an immediate fail-loud error - that env var is mandatory, and it must match a user in users.json. Second: a fresh install has zero enabled backends, so your first submit will tell you to configure backends.json. That's the pack refusing to phone random hosts. Third: prompt_json must be genuine API-format JSON - paste a raw workflow you saved from the UI's "Save (API Format)" button, not a PNG. And a genuinely funny one the pack's own config warns about: the local-selftest backend loops back to this ComfyUI, so a blocking self-submit waits behind itself in its own queue. Keep that one on wait_for_completion=false.