☁️BizyAir Seedance 1.0 Pro Text To Video

What it is

This node hands your prompt to Seedance, ByteDance's video model, and gets a clip back - no local GPU involved, because the GPU isn't yours, it's ByteDance's. Seedance is the video half of ByteDance's closed generation lineup (Seedream does images), served here through Doubao, the same consumer AI platform ByteDance uses domestically. It has never shipped as open weights and never will - the KB's own read on the split is blunt: "Vidu, Kling, Hailuo, all closed source," and Seedance sits right there with them. If you want ByteDance's video quality, an API call like this one is the only door in.

That's the trade you're making by using this node: you get a genuinely strong closed model without touching a training run or a checkpoint file, but you're paying per generation through your BizyAir account and you have zero control over what's under the hood.

How it works

You write a prompt, pick a model variant, and BizyAir round-trips the request to Doubao's Seedance endpoint. The two model choices are doubao-seedance-1-0-pro-250528 (the dated pro release) and doubao-seedance-1-0-pro-fast-251015 (a faster variant, presumably trading a bit of quality for turnaround - the node doesn't say more, and neither does the README, so treat "fast" as exactly that and nothing more specific). Whichever you pick, the job runs entirely server-side; ComfyUI is just the front end submitting the request and waiting for the video to land.

Inputs and outputs that matter

prompt is the only thing you're really authoring. resolution (480p/720p/1080p) and ratio (16:9 down to 21:9, plus adaptive) jointly decide the actual pixel dimensions - the tooltip is explicit that resolution and ratio combine to set the frame size, and for the exact numbers you'll need Doubao's own docs, not the node. duration runs 2–12 seconds. fps is locked to 24, so it's really just there for visibility. camerafixed asks the model to hold the camera still by appending an instruction to your prompt - the tooltip's own caveat is that the effect "isn't guaranteed," so don't build a workflow that depends on it. seed defaults to -1 (random); pin it if you need to reproduce a result.

Output is video, which is what you'll save or chain into further processing, plus bizyair_model_name - an internal tag identifying which underlying model actually served the request. It's there for logging and debugging multi-model workflows more than for wiring anywhere.

Installing it

Easiest path: open ComfyUI Manager, search "BizyAir," hit install, restart. Manually, it's cd ComfyUI/custom_nodes && git clone https://github.com/siliconflow/BizyAir.git then a restart. Either way, you need a BizyAir API key before anything works - first run, click the "click to login" prompt inside ComfyUI and register at bizyair.cn / siliconflow. Without a key, every node in this pack - this one included - fails immediately.

Common issues

The failure mode you'll hit most is a rejected or empty request because the account isn't authenticated or is out of credits - this is a metered API call, not a free local node, so check your BizyAir balance before assuming the node is broken. Beyond that, watch duration: it's capped at 12 seconds, so if you're expecting a longer clip you'll need to generate multiple segments and stitch them, there's no way to push past the ceiling from inside ComfyUI. And because Seedance never publishes model weights, there's no local fallback if the API is slow or down - that's the whole bargain with any node in this "External APIs" category.