Seedance 2.5 Is Live: 30-Second Video and 50 References

Seedance 2.5 is live in ByteDance's Dreamina products and on Yix. This guide separates the model's full public capabilities from the workflows that Yix supports today.

ByteDance first showed the model at its FORCE conference on June 23, 2026. The pitch focused on a practical problem with AI video: five or ten seconds is often enough for a striking clip, but not enough for a scene with a setup, movement, and an ending. Seedance 2.5 stretches a single generation to 30 seconds and gives the model far more reference material to work with.

That sounds like a version-number update. It is closer to a change in how a shot can be planned.

What ByteDance announced

36Kr's report from the FORCE event lists three changes that matter for production work:

• A single generated video can run for up to 30 seconds.
• One job can use as many as 50 multimodal reference assets.
• The model can make targeted changes to part of a clip instead of forcing a complete rerender.

The reference limit is the interesting part. A creator can give the model character images, a product shot, movement footage, audio, a storyboard, and visual direction in the same job. Earlier workflows often squeezed those decisions into a prompt or split them across several tools.

ByteDance also demonstrated reference-to-video control with green-screen footage and simple 3D "white model" previs. Those rough scenes carry blocking, camera movement, and spatial relationships from preproduction into the generated shot; they are not meant to be finished 3D renders.

Why 30 seconds matters

Thirty seconds is still short by normal filmmaking standards. For AI video, it is enough room to expose whether a model understands continuity.

A five-second clip can hide a lot. The camera moves, something dramatic happens, and the video ends before the scene has time to drift. A 30-second shot has to remember where people are standing, what they are wearing, how the light falls, and what action should happen next. More time also makes timing prompts useful. You can ask for a turn at six seconds, a reveal at twelve, and an exit near the end instead of hoping the model invents a sensible rhythm.

Dreamina's official Seedance 2.5 page also advertises 4K output, synchronized audio, multilingual prompting, local edits, and a beta long-video mode. Some accounts and regions are receiving features before others, so the product page and the model selector may not agree yet. That is normal for a staged rollout, but it is worth checking before you plan a deadline around one specific feature.

What changed from Seedance 2.0

Seedance 2.0 already combined text, images, video, and audio in one generation. It also handled native sound, video editing, and video extension. Version 2.5 pushes the same idea further:

• The standard shot is longer.
• The reference budget is much larger.
• Previs and motion references have a clearer role in directing the result.
• Local edits make it easier to fix a prop, person, or background without discarding everything else.

A product ad can keep a fixed pack shot. A short scene can reuse two characters. A previs pass can ask the camera to follow an existing move. The model still has to prove how reliably it can do this across repeated runs. A feature limit tells you what the interface accepts, not how many outputs you will actually keep.

What the full model supports

The complete Seedance 2.5 workflow can combine text, images, video, audio, and script references. It supports clips up to 30 seconds, native synchronized audio, reference-guided generation, targeted editing, and continuation. Some products expose only part of that surface, so check the available controls before planning a production workflow.

How Seedance 2.5 works on Yix

Select Seedance 2.5 in the Yix AI video generator for text-to-video or add one starting image for first-frame image-to-video. The current Yix interface supports 480p and 720p output, synchronized audio, and durations from 5 to 30 seconds.

The current Yix workflow deliberately exposes a focused subset of the full model. It does not yet accept reference videos, reference audio, multi-image reference packs, editing jobs, or extension jobs.

For the supported text and starting-image workflows, Seedance 2.5 starts at 380 credits for 5 seconds at 480p. The default 5-second 720p setting costs 860 credits. The generator shows the exact Yix credit total before charging.

Plan a UI product promo with it

If your starting asset is a landing-page mockup or SaaS dashboard, use the UI mockup to product promo workflow. It includes a 15-second shot plan, a copy-ready prompt, and the important boundary between a generative concept film and a real screen-recorded product demo.

You can also load the UI Product Promo preset in the Prompt Lab, copy the production brief, and run it with Seedance 2.5 in the video generator.

What you can use on Yix today

The Yix AI video generator now offers Seedance 1.5 Pro and Seedance 2.5. Seedance 1.5 Pro is the lower-cost option with 480p through 1080p output and 5–12 second clips. Seedance 2.5 supports longer 5–30 second clips at 480p or 720p, stronger prompt following, and native synchronized audio.

Yix prices Seedance 2.5 by output duration and resolution. Audio currently does not change the Yix credit cost for this model, and the exact total is shown before generation.

Sources: Dreamina's official Seedance 2.5 page and 36Kr's FORCE event report.

Seedance, Dreamina, and ByteDance are trademarks of their respective owners. Yix is not affiliated with them.