Skip to main content

​
Local ComfyUI setup

​
What this lets you do

Run your own ComfyUI workflow through DreamLayer WebUI, the SDKs, or an orchestrator.

​
What you need

  • A local ComfyUI installation
  • Every checkpoint, LoRA, and custom node required by the workflow
  • A DreamLayer clone running on the same machine

​
Steps

​
1. Start ComfyUI

Run ComfyUI on its loopback address. DreamLayer does not install ComfyUI, models, checkpoints, LoRAs, or custom nodes.

​
2. Export API-format JSON

In ComfyUI, use the developer option to save the workflow in API format. The normal browser workflow format is not executable through the transport.

​
3. Create the runtime configuration

Copy the shipped example, then set the workflow path, output node, supported operations, and any DreamLayer placeholders used by your graph.
Use __DREAMLAYER_PROMPT__, __DREAMLAYER_WIDTH__, and __DREAMLAYER_HEIGHT__ in the exported graph. Add __DREAMLAYER_INPUT_IMAGE__ only when the workflow declares image_to_image.

​
4. Save the configuration and validate

When prompted, enter the absolute path to comfyui-config.json.

​
5. Map inputs and output

The workflow registration declares which node inputs accept prompt, width, height, and an optional reference image, plus which node produces the final image.

​
Confirm it worked

./dreamlayer doctor reports the ComfyUI configuration as ready, the configured workflow appears in DreamLayer WebUI, and one turn returns the mapped final image.

​
Common errors

An unavailable checkpoint, custom node, input mapping, or output mapping fails visibly before DreamLayer reports a completed result. No silent substitution occurs.

​
Troubleshooting

Confirm that ComfyUI answers at http://127.0.0.1:8188, the file is API-format JSON, and all referenced nodes are installed. DreamLayer does not validate every model and hardware combination.

​
Next steps