No extra Python dependencies. The nodes use the standard library plus the torch / numpy / Pillow that ComfyUI already ships, so installing them cannot break your environment.

| Node | What it does | | --- | --- | | APIMaster Config | Endpoint + key, passed to the other nodes | | APIMaster Text to Image | gpt-image-2, doubao-seedream-5-0-pro-260628, gemini-3.1-flash-image, Midjourney → IMAGE | | APIMaster Image to Image | Reference images (and an optional mask) → IMAGE | | APIMaster Video | seedance, kling, MiniMax-H3, grok-imagine-video → an MP4 in your output folder | | APIMaster List Models | Prints what the endpoint actually serves right now |

Install

ComfyUI Manager

Search for APIMaster in ComfyUI Manager and install. Restart ComfyUI.

Manual

cd ComfyUI/custom_nodes
git clone https://github.com/apimaster-ai/ComfyUI-APIMaster
# nothing to pip install

Restart ComfyUI.

Set your key

Paste your key into the api_key field of the APIMaster Config node.

Clear it before you share. Widget values are saved inside the workflow JSON, and workflow JSON gets shared — posted in issues, embedded in generated PNGs, published on model sites. Empty the api_key field before sharing a workflow or an image made with it.

The node reads the key from that field only. It does not read the APIMASTER_API_KEY environment variable or ~/.apimaster/config.json (the file apimaster-cli saves its key in): custom nodes on the Comfy Registry may not read credentials from the environment or files outside ComfyUI. Versions up to 0.1.4 read the variable, and up to 0.1.3 the file.

Example workflows

Drag either file from example_workflows/ onto the ComfyUI canvas:

text_to_image.json — prompt → gpt-image-2 → preview
image_to_video.json — load an image → seedance-2.5 → MP4

Notes that will save you time

Sync vs async for images. Sync mode returns the image in one request and is simpler. For 2k and especially 4k it can exceed the gateway's own timeout and come back as a 408 — switch the node's mode to async and it submits the job and polls instead.

Aspect ratio in image-to-video. Always set aspect_ratio explicitly. A portrait reference image with no aspect set is treated as 16:9 by the gateway, and you get a letterboxed result.

Video is slow. seedance-2.5 took about 15 minutes for a 4-second clip. The node waits up to 30 minutes and keeps polling through brief network drops; 720p is the safe resolution, higher tiers depend on the model.

Advanced image parameters change the price. Quality, background and output-format options narrow which upstream channels can serve a request, which can route you to a more expensive one. The nodes send only the fields you actually set.

Batch sizes. If the endpoint returns images of different sizes in one batch, the node raises instead of silently resizing. Set n=1 or pass an explicit size.

Which models are available

Model catalogs change. Drop an APIMaster List Models node into any workflow, set filter to image or video, and run it — the list is printed to the node and to the console. Or from a terminal:

npx @apimaster/cli models --kind image

Use it with another provider

The nodes are not hardcoded to APIMaster. Point base_url at any OpenAI-compatible endpoint that implements /images/generations, and use the model_override field for a model id that is not in the dropdown.

Paste that endpoint's key into the config node. When you load a workflow someone else shared, check its base_url before typing your key in: the key goes to whatever server that field names.

Development

python -m unittest discover -s tests -v

The tests run against a local mock server built into the repo — no key, no network, no tokens spent.

Four of them exercise the ComfyUI IMAGE contract (tensor → data URI → bytes → tensor) and need torch, so they skip on a plain Python install. Run them inside ComfyUI's own environment for full coverage:

# from ComfyUI/custom_nodes/ComfyUI-APIMaster
../../.venv/Scripts/python.exe -m unittest discover -s tests    # Windows
../../venv/bin/python -m unittest discover -s tests             # macOS / Linux

Verified on ComfyUI with all five nodes appearing in /object_info, and both example workflows validated against that live node registry.

Troubleshooting

| Symptom | Cause | | --- | --- | | HTTP 401 | Key has stray whitespace or quotes, or the api_key field is empty | | HTTP 404 | base_url is missing the /v1 suffix | | HTTP 400 ... use the Images API | A chat model id was selected for an image node | | HTTP 402 | Out of balance | | HTTP 408 | Sync generation timed out — switch to async | | Nodes missing after install | Restart ComfyUI; check the console for an import error |

License

MIT