H3 Imagine

MiniMax H3-Image ComfyUI Workflow (Download + Setup Guide)

This guide covers a practical H3-Image workflow: setup, API configuration, first generation, and the settings most likely to need adjustment. It is for people who want a ComfyUI graph after the browser workflow is already familiar.

Workflow download

Add your exported workflow JSON to this repository before publishing. The download control is intentionally disabled until a tested H3-Image node and its version are pinned.

Before you start

Use a current ComfyUI build, a modern browser, and a MiniMax developer account with an image API key. Because this workflow calls a hosted image API, GPU memory is mostly relevant to the rest of your ComfyUI setup rather than the remote generation itself. Keep the key outside any file you intend to share.

Setup in eight steps

  1. 1. Install a current ComfyUI build. Use the portable or desktop release and open the Manager panel once it is running.
  2. 2. Install the H3-Image custom node. Search the Manager for the verified H3-Image integration supplied by its maintainer; do not install a similarly named unverified node.
  3. 3. Create a MiniMax API key. Create a least-privilege key in the official MiniMax platform and record its endpoint configuration.
  4. 4. Add the API key to the node configuration. Enter it through the node credential field or a local environment variable so it never appears in an exported JSON.
  5. 5. Download the workflow JSON. Use the tested JSON linked above when it becomes available, rather than rebuilding parameter names from screenshots.
  6. 6. Load the workflow in ComfyUI. Drag the JSON into the ComfyUI canvas and install any missing dependencies reported by the loader.
  7. 7. Set prompt and reference images. Set the prompt, choose an aspect ratio, and connect only the reference images that matter to the requested result.
  8. 8. Queue the first run at 1K. Start with one 1K image to validate the connection, then raise size or count after the API response is correct.

Common errors

Model loading fails

Confirm that the node version matches your ComfyUI release, then restart after installation.

Out of memory

Reduce output size and batch size first. Cloud API workflows do not need a large local diffusion checkpoint.

Missing node

Use ComfyUI Manager to install the custom node named by the workflow, then refresh the browser.

Black output

Check that the outgoing API request has a valid prompt and image URL or supported image input.

Sampler error

Remove sampler-only fields: an API-backed image node may not accept local diffusion sampler settings.

API key is not configured

Store the key as a local environment variable or in the node setting. Never save it in a public workflow file.

Parameters worth adjusting

Begin with a single output and 1K resolution. Use a direct prompt before adding references. For reference images, include only the source material that controls identity, product geometry, or composition. If an integration exposes reference weight, begin at a middle value: increase it when subject identity drifts, and lower it when the result needs more freedom. The official API, rather than a local sampler, controls its own image-generation behavior.