ByteDance Seedance 2.5 Text to Video

ByteDance2TextToVideoNode is the pure text-to-video member of the Seedance 2.0 family: prompt in, VIDEO out, no image anchors, no reference assets. It's a built-in partner node, so that's a cloud API call billed through your Comfy account - the node uploads your prompt to Comfy's API layer, ByteDance's Seedance 2.0/2.5 models render it, and the finished clip lands back on your disk. Nothing runs on your GPU, and there's no vendor key to paste; being logged into Comfy is the whole auth story.

The reason you'd pick the 2.x node over the older ByteDance Text to Video is the model family and its controls. The 2.x series is ByteDance's current flagship line, and the newer 2.5 tier adds a genuinely useful trick: put spoken lines in double quotes in the prompt and the model steers the dialogue it generates. That's rare at this quality level - most video APIs ignore speech entirely.

The inputs

The layout surprises people the first time: the model dropdown is most of the UI. Pick Seedance 2.0, 2.0 Fast, 2.0 Mini, or 2.5 and the node exposes the per-model options underneath - prompt, resolution, ratio, duration, and an audio toggle. So don't hunt for a separate prompt box at the top level; it lives inside your model selection.

• model - quality/speed/cost ladder: 2.0 for max quality, 2.0 Fast for speed, 2.0 Mini for the fastest, lowest-cost generation.
• prompt (inside the model) - your text description. Quote dialogue to steer spoken lines on 2.5.
• seed - the usual: it only forces a re-run; results are nondeterministic regardless of seed.
• watermark - add an "AI generated" watermark or not.

What comes out

One VIDEO output. Save it, preview it, or chain it into an edit/extend node.

Gotchas

• The prompt is inside the model dropdown. If your text box "disappeared," you changed models - re-pick and it comes back. This is the #1 confusion with these dynamic-combo nodes.
• Dialogue steering is a 2.5 feature. Don't expect quote-steered speech on the 2.0 tiers.
• Model choice is a real cost lever. Fast/Mini exist precisely because a 2.0-quality render is slower and pricier; use them for drafts, then spend the credits on the final.
• Seed ≠ reproducibility. Same disclaimer as every node in this family: it determines re-running, not the outcome.

The whole family pattern is consistent - pick a model, write a prompt, pay per call, save the video. This node is just the one with no training wheels (and no image) between you and a Seedance render.