ByteDance Seedream V4 (Official Edition)

A text prompt, a size, and a finished image comes back - ByteDance's Seedream V4 text-to-image model running in the cloud, wrapped as a ComfyUI node by BizyAirPlus. There's no checkpoint to download and no GPU needed, because Seedream is API-only; this node is the official route into it.

Seedream sits at the sharp end of ByteDance's closed image line. The open ecosystem gets its components, but the flagship generators stay behind an API - and this node is the ComfyUI-native way to use one of the strongest text-to-image models available without leaving your canvas. It's the right pick when prompt fidelity matters more than local control: you trade per-image billing for a model that consistently follows instructions better than most open checkpoints. The two-input surface (prompt + size) tells you this node is a pure generator, not a tuner - that's by design.

How it works

Flip BizyAirPlus ON in the action bar and this output node sends your prompt and size to Seedream's service through the vendor's direct API, then pulls the finished image back into your workflow. The Official Edition label is about the route - the same model is reachable through BizyAir's channel nodes too, and which one you pick comes down to billing and availability on your account.

The inputs that matter

• prompt - multiline, free-form. Seedream rewards descriptive natural language.
• size - a string like 2048x2048 (the default) giving the exact output dimensions in pixels. Unlike the combo boxes on some sibling nodes, here you type the numbers.

The optional pair is worth knowing: api_config overrides the base URL and API key for a single request (proxy setups), and skip_error makes the node emit an error placeholder instead of aborting your whole workflow when a request fails - handy for unattended batches.

The outputs

image is an IMAGE tensor you can wire into anything downstream - save it, feed it to an upscaler, use it as a reference elsewhere. urls gives the direct result link, and request_id is the ID to keep if you need support to look at a failed generation.

Installing it

This node ships in BizyAirPlus. Install via ComfyUI Manager (search "BizyAirPlus") or:

cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt

Restart ComfyUI, turn the BizyAirPlus button ON, and enter your API key from bizyair.ai when prompted. The key is saved to ~/.BizyAirPlus/apikey.ini, or set BIZYAIR_API_KEY as an environment variable. The real node code lives in the bizytrd / bizyair-cloudberry pip packages - the install is light and there are no model downloads.

Gotchas

Nearly every first-run failure is the button being OFF or a missing/stale key; the ComfyUI console has the actual error. One thing worth saying clearly, because it keeps circulating: BizyAir is not malware - the "suspicious extension" reputation came from older ComfyUI Manager builds auto-installing it as a missing-node suggestion, a registry inference bug that's since been fixed. On this node specifically, watch the size field: it's a plain string, so typos like 2048 x 2048 won't parse the way you expect. And if your final image needs more than a straight text-to-image pass, remember Seedream V4's Edit sibling exists in the same pack.