Seedance 2.5 is now live on film.fun.
ByteDance's latest video model is wired into Splice as seedance25 on the Replicate path — pick it anywhere you generate video: text-to-video, image-to-video, and the agent's Shot Sheet loop. It's the longest, most audio-complete model in the lineup.
What's new vs Seedance 2
Seedance 2.5 keeps the unified audiovisual architecture — audio is generated in the same pass as the picture, not dubbed on top — and pushes the envelope on length and control:
- Shots up to 30 seconds. Where most models tap out at 5–10s, 2.5 will hold a single coherent take far longer — enough for a full beat, not just a cutaway.
- Native synchronized audio. Dialogue, foley, ambience, and score come out of the same generation. Put spoken lines in "double quotes" in your prompt and the model lip-syncs them.
- First- and last-frame control. Pin the opening image, the closing image, or both, and the model tweens a believable motion path between them — the cleanest way to hit an exact start and end.
- Full reference stack. Reference images for character and style continuity, reference videos as motion priors, and reference audios for voice and musical tone. Mix them to lock a look, a move, and a sound in one shot.
- 480p and 720p, with adaptive aspect ratios (16:9, 9:16, 1:1, and more) so multi-platform delivery is one generation, not three.
How to prompt it
- Lead with the shot, not the story. "Slow dolly-in on a rain-streaked window, neon bleeding through, shallow depth of field" beats a paragraph of plot.
- Quote your dialogue. Anything in "double quotes" is treated as spoken audio and lip-synced. Everything else describes picture and sound design.
- Keep it under 2,000 characters. Seedance 2.5 hard-caps the prompt — a very long, over-augmented prompt will fail validation. Splice truncates on a word boundary, but tighter prompts generate better anyway.
- Reach for references before adjectives. For a recurring character or a specific motion, a reference image or video does more than a sentence ever will.
When to reach for it
Choose Seedance 2.5 when you need a long, audio-complete take — a talking shot, a scene with its own soundtrack, or a start-to-finish move you want to nail with first/last frames. It's the model to beat for length and native sound.
One practical note: at peak times Seedance 2.5 can sit in a busy queue on Replicate, so a long shot may take a while to come back. Kick it off, keep working, and it'll land in your Library when it's ready.
Priced per second of output, billed only for what you generate. Pick Seedance 2.5 in Create → Video and start with a short test shot before committing to the full 30 seconds.
