Connect ComfyUI
The media bridge runs beside ComfyUI and can execute exact Grid image or video recipes. Having ComfyUI installed is only the first state; it does not make an arbitrary checkpoint or custom graph a public Grid capability.
Install or Start ComfyUI
Keep the ComfyUI installation and custom nodes you already operate. New users should follow the official ComfyUI installation guide for their platform. Start ComfyUI independently; the Grid’s browser UI does not run remote shell installers or accept a browser-selected executable.
Detect the Endpoint
The common local API is http://127.0.0.1:8188. The media bridge checks the
running API and inventories local weight filenames. Detection is descriptive
only and never enrolls the machine.
Capability States
| State | Meaning |
|---|---|
| Detected | The local ComfyUI API reports model-weight files |
| Compatible | A bundled Grid workflow resolves every required local weight |
| Qualified | The approved profile or review gate and canary have passed |
| Advertised | The running worker registered that capability with Core |
These states are intentionally separate. Compatibility never grants qualification or silently advertises a model.
Compatible Models and Recipes
A capability is compatible only when a known Grid workflow resolves and every weight referenced by that graph exists locally. Similar filenames and arbitrary custom workflows do not qualify. Recipe-executor jobs use the exact recipe Core resolves; local mapping remains a restricted fallback.
Inspect and Configure a Candidate
- Start ComfyUI. Its common loopback endpoint is
http://127.0.0.1:8188. - Follow the current media path on aipowergrid.io/run.
- In the local bridge UI, test the endpoint and refresh the capability inventory.
- Review detected weights and known compatible recipes.
- Select only the capabilities you intend to expose.
- Complete the required qualification and canary before advertising.
- Once a signed serving worker is online, run the manager’s separate Grid test to verify exact-worker dispatch and output transport. It is economically inert setup evidence, not a quality or fidelity verdict.
Public managed-media onboarding is qualification-gated. Do not work around that gate by entering a model name manually or setting a trust override.
Media workers currently accept one Grid render at a time. They support pause, maintenance windows, and drain-before-disconnect. Grid work and local ComfyUI work still share the same queue; a configured schedule is not universal idle queue detection.
Security Boundary
Input images, prompts, and outputs are plaintext to the community worker. Model files remain local, and uploads use short-lived Grid URLs rather than storage credentials.
The bridge connects outbound to the Grid and defaults its ComfyUI and management surfaces to loopback. It never requests a payout-wallet private key. Remote management should use an SSH tunnel rather than exposing the local UI.
Verify Qualification
Run the exact downloaded qualification artifact and submit the documented privacy-reduced evidence pair. The public status on aipowergrid.io/run is the authoritative release and evidence gate. Until a signed profile is published, there is no public worker-registry entry to verify and no reward eligibility.
Troubleshooting
- ComfyUI not detected: confirm its API is listening on loopback port
8188or enter the operator-controlled URL in the local bridge UI. - Detected but no compatible capability: compare the exact required weight names and workflow dependencies; do not force a model name override.
- Canary fails: keep the machine offline from Grid serving and attach only the sanitized qualification report requested by the cohort instructions.
- Local renders queue behind Grid work: use pause or maintenance windows. Automatic ComfyUI queue-idle yielding is not implemented.