ByteDance Seedance 2.0 Mini Text to Video (Channel Edition)

This node turns a text prompt into a video using ByteDance's Seedance 2.0 Mini - without downloading a single weight file or lighting up your GPU. That's the whole pitch, and it's a good one. Seedance is one of the best closed video models around, but it's API-only; this node is the ComfyUI-shaped door to it, served through SiliconFlow's BizyAirPlus cloud.

If you've ever queued a local video model and watched your GPU thermometer spike while a 20-second clip takes an hour, you get why this exists. The trade is money instead of VRAM: you need a BizyAir API key and each request bills against your account, but the generation itself happens on their servers and the clip comes back to your workflow as a normal video file.

How it works

BizyAirPlus is a cloud-execution layer for ComfyUI. You build the workflow as usual, then flip the BizyAirPlus ON button in the action bar so the cloud nodes take over. This one is an output node: it packages your prompt, resolution and duration into an API request to ByteDance's Seedance service, waits for the cloud to render, and hands the finished video back to ComfyUI. Under the hood the node itself is thin - the real machinery lives in the bizytrd Python package that installs alongside the pack. The "Mini" in the name is the cheaper, faster tier of Seedance; the "Base" / "Channel Edition" is BizyAir's route to it.

The inputs that matter

• prompt - the text prompt describing the video. This is where you spend your effort; Seedance is responsive to natural-language scene descriptions.
• resolution - output video resolution, defaults to 480p. Mini is a budget tier, so don't expect 4K out of it.
• duration - seconds of video, 4 to 15 (default 8).
• aspect_ratio - set to something other than auto when you know the target, like 16:9 for YouTube or 9:16 for shorts.
• seed - -1 means fully random; set a fixed value to make a generation reproducible.
• generate_audio - on by default, so you get a soundtrack baked in rather than a silent clip.
• watermark, return_last_frame, web_search - all off by default. web_search feeds current context into prompt enhancement, which is occasionally magic and occasionally nonsense; leave it off until you have a reason.

There's also api_config (per-request override of the base URL and API key - useful if you're routing through a proxy) and skip_error, which makes the node emit an error placeholder instead of killing your workflow when a request fails. Nice for batch runs.

Wiring the outputs

You get three: video, request_id, and urls. Wire video into a preview node or a Video Combine / save node to keep the clip. urls is the direct link to the result if you want to grab the file yourself. request_id is the one beginners ignore and then need: when a generation goes wrong, BizyAir support wants that ID.

Installing it

This node ships in BizyAirPlus, not as a standalone package. In ComfyUI Manager, search "BizyAirPlus" and install, or:

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

Then restart ComfyUI. The requirements pull in bizyair-cloudberry and bizytrd from PyPI - no model files, nothing heavy. First run, flip the BizyAirPlus button ON and it'll prompt for your API key from bizyair.ai. The key lives in ~/.BizyAirPlus/apikey.ini, or set BIZYAIR_API_KEY as an environment variable.

Gotchas

BizyAir has a slightly unfair reputation as sketchy software, because early on the Manager would auto-install it as a "missing node" whenever a workflow using cloud nodes was loaded - which read as malware to a lot of people. It's not; it's a legitimate SiliconFlow service, and the registry inference that caused the confusion was fixed. The real gotchas are yours: forget to turn the button ON and the node errors, and cloud failures always trace back to a bad or missing key. When in doubt, check the ComfyUI console - the detailed error is always there.