ComfyUI-APIMaster

Image and video generation nodes for ComfyUI, backed by APIMaster or any other OpenAI-compatible gateway.

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

Set an environment variable before starting ComfyUI:

# macOS / Linux
export APIMASTER_API_KEY=sk-...

# Windows PowerShell
$env:APIMASTER_API_KEY = "sk-..."

Then leave the api_key field in the Config node empty.

Why not just type it into the node? Because widget values are saved inside the workflow JSON, and workflow JSON gets shared — posted in issues, uploaded with generated images, published on model sites. Keys leak that way constantly. The node prints a warning if you fill the field in anyway.

The key is also read from ~/.apimaster/config.json if you use the apimaster-cli.

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.

For any base_url other than https://apimaster.ai, paste that endpoint's key into the config node. APIMASTER_API_KEY and ~/.apimaster/config.json are only ever sent to apimaster.ai: base_url is part of the workflow, so a workflow someone shares with you could otherwise point it at their own server and receive your key.

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 it was never set in the environment ComfyUI runs in
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