BytePlus Seedance 2.5 Text to Video (Direct API)

If you found this page because an old workflow you saved still references this node, relax: nothing is broken. BytePlus Seedance 2.5 Text to Video (Direct API) is the earlier, text-only node for calling ByteDance's closed Seedance 2.5 model through the BytePlus ModelArk Direct API. The maintainer has since folded everything it did into the unified BytePlus Seedance Reference to Video node and marked this one deprecated - it's hidden from node search by default and kept only so existing workflows keep opening and running.

So the short version of this article: if you're building a new graph, add the Reference node instead and leave its reference sockets empty. You get identical text-to-video plus the option to wire in images, audio, or a source clip later, and you get the model selector (this one is hard-wired to Seedance 2.5). If you're reopening a canvas that already uses this node, it still works - the pack registers it, just quietly.

What it does

Type a prompt, pick your settings, and the node submits a paid generation task to the ModelArk endpoint using the ARK_API_KEY on your BytePlus account. No weights, no local GPU - Seedance 2.5 only exists behind ByteDance's API, so you're renting a remote render and waiting for it. When the job finishes, the node downloads the clip and hands it out of the VIDEO output, which you feed to the core Save Video node.

Because this is a real paid task, the billing cautions apply: queueing can charge your account, task creation is never auto-retried (a duplicate would mean a duplicate charge), and stopping ComfyUI doesn't cancel a job that's already been created. If an error shows after the task was created, it includes a task ID - paste it into resume_task_id to continue that same generation instead of paying for a new one.

The inputs

None of this is exotic - it's a trimmed-down version of the Reference node:

• prompt - multiline text describing the shot.
• resolution - 480p, 720p, or 1080p; 720p is the default and the sensible place to start.
• ratio - 16:9 by default, with 4:3, 1:1, 3:4, 9:16, 21:9, and adaptive available.
• duration - 4 to 30 seconds, default 5.
• generate_audio - on by default; Seedance 2.5 returns a clip with a matching audio track.
• watermark, output_format (mp4 or mov) - advanced niceties you can leave alone.

Two inputs people misunderstand: rerun_seed and endpoint_id. The seed is never sent to Seedance 2.5 - it only busts ComfyUI's cache so the node will actually re-queue, so don't expect to reproduce a clip by fixing it. And endpoint_id is only for keys restricted to a specific ModelArk inference endpoint; it's saved in the workflow, so it's not a place for an ark-... secret.

Install and key

It's the same pack as the Reference node, so install is shared: Manager → search BytePlus Seedance Direct API, or

cd ComfyUI/custom_nodes
git clone https://github.com/raucvr/byteplus-seedance-direct.git
cd byteplus-seedance-direct
python -m pip install -r requirements.txt

You need ComfyUI 0.34+, and the only Python dependency is truststore - no model downloads. Then configure the key: in-app on macOS/Windows ComfyUI Desktop (stored in the Keychain or Credential Manager), or an ARK_API_KEY environment variable on Linux/WSL. Full details live on the BytePlus Seedance Reference to Video page for this pack.

One install gotcha specific to old workflows: if you previously installed the standalone version under the folder name byteplus_seedance25, disable it before installing the Registry version. Keeping both copies registers the same node IDs twice, and the "missing node" errors you get on load are usually that, not the deprecation. Otherwise, an old canvas loads, the node runs, and you keep generating Seedance 2.5 clips without touching anything.