🌱 Seedance 2.0 Text-to-Video

Seedance 2.0 Text-to-Video is the entry point to this whole pack: type a prompt, wire in your BytePlus API key and endpoint, queue, and walk away while ByteDance's closed Seedance 2.0 model renders a clip on their servers. There's no checkpoint to download, no VAE to argue with, no VRAM to run out of - the entire pipeline lives in the cloud and this node is just the HTTP client that talks to it.

That's the thing worth internalizing before you use it. Seedance 2.0 is a flagship ByteDance model and it is API-only - you cannot download it, so an API node is the only door into it. What you get in exchange is a surprisingly full control set and a video URL at the end. What you pay is per-call cost (Seedance renders are not cheap, and the community reacted to them arriving in ComfyUI with genuine sticker shock), plus your prompt leaving the machine to a server with ByteDance's content moderation bolted on.

The inputs that matter

The default prompt is a cinematic aerial shot of a futuristic city at dusk - a decent showcase, but you'll replace it. Beyond the prompt:

• aspect_ratio - smart, 16:9, 9:16, 4:3, 3:4, 21:9, 1:1. Defaults to 16:9.
• resolution - 480p, 720p, 1080p, 4k. Tooltip warning you need to read: 1080p/4k availability depends on your endpoint. Your ep-... deployment may simply not offer 4K, and you'll find out via an API error.
• duration - 4–15 seconds, 1-second steps, default 5.
• seed - -1 means random; any other value for reproducibility.
• batch_count - 1–10. The tooltip says it plainly: "Create multiple tasks with different seeds. Cost and wait time scale with this count." This is the node's hidden money-burner. Ten clips at 4K 15s is a real invoice.

Optional inputs are api_key, endpoint, and generate_audio (default true - Seedance can produce synchronized sound, not just video).

How it works, and the outputs

Under the hood it POSTs a task to BytePlus (/api/v3/contents/generations/tasks), polls until it succeeds, and hands you four outputs: video_url (the STRING you'll preview or save), first_frame (the first frame decoded as a ComfyUI IMAGE, handy for a quick thumbnail), request_id (your cgt-... task ID for later retrieval), and batch_json (a JSON summary of every task in the batch, for the batch browser nodes).

Realistic wait times: roughly 5–25 minutes for a single clip, and it gets worse during China peak hours. The queue is the experience - this node is not a sampler, it's a job submission.

Install, and the one thing everyone forgets

ComfyUI Manager → Install via Git URL → https://github.com/nnnnkatsu/seedance2-comfyui-byteplus, restart. No model downloads.

The thing everyone forgets: the video URL expires. BytePlus keeps task data and output URLs for roughly 24 hours, then they're gone. If you render something you like, wire video_url into Seedance 2.0 Save Video immediately - that node downloads the mp4 to your output folder before the link rots. A render you don't save is a render you eventually can't find. And if it's a real person in the output, remember the moderation runs on ByteDance's terms, not yours.