Seedance 2.5 is Live on Runable: 30s One-Take AI Video

Key Takeaways

• Seedance 2.5 is live on Runable from day one, bringing 30 second one-take video with synchronized audio, multi-minute extensions, and reference-driven editing to every Pro and Max workspace.
• A single Seedance 2.5 pass generates up to 30 seconds of video and audio together, double the 15 second ceiling of the previous generation, and multi-round extension chains passes into multi-minute videos.
• One generation accepts up to 50 multimodal references: 30 images, 10 video clips, and 10 audio clips, so characters, locations, camera language, and sound are locked before you generate.
• Timestamp-level editing changes a specific moment inside a finished clip without regenerating the whole thing, alongside green screen, camera perspective, and clay render editing workflows.
• Runable pricing as of August 2026: a Free tier, Pro at $20/mo (most popular), and Max at $100/mo. Seedance 2.5 is included on Pro and Max.

Seedance 2.5 Launches on Runable: What's New and How to Use It

We're pleased to announce the release of ByteDance's Seedance 2.5, now available on Runable from day zero.

Seedance 2.5, announced by the ByteDance Seed team on July 31, 2026, is built around two ideas: one-take creation and flexible referencing. Instead of stitching together short silent clips and layering sound on afterwards, the model generates picture and audio together in a single pass. And instead of describing everything in prompt text, it takes direction from the material you already have: character sheets, location stills, motion references, and voice clips.

The result is a model that behaves less like a clip generator and more like a small production unit. You brief it with references, it delivers a continuous take with sound, and you correct individual moments instead of rerolling the whole scene.

This guide walks through everything new in Seedance 2.5 and how to use it on Runable.

Seedance 2.5 Key Strengths

With Seedance 2.5, Runable users get:

• Up to 30 seconds of video with synchronized audio in a single generation, double the previous 15 second limit
• Multi-round extension, which continues a finished clip into multi-minute videos while preserving characters, style, and sound
• Up to 50 multimodal references in one pass: 30 images, 10 video clips, and 10 audio clips
• Timestamp-level editing, so you can fix one moment without regenerating the rest
• Editing workflows that used to require a studio: green screen compositing, camera perspective changes, and reference-based scene edits

This is a real shift in how you direct AI video. Instead of gambling an entire scene on one paragraph of prompt text, you hand the model your references up front and fix individual seconds after the fact.

Model Improvements

Seedance 2.5 targets the exact problems that kept earlier AI video out of real productions: clips that were too short, output that was silent, characters that drifted between generations, and no way to fix a small mistake without starting over. Each of the upgrades below removes one of those walls.

30 Second One-Take Storytelling

Seedance 2.5 generates up to 30 seconds of continuous video with synchronized audio in a single pass, twice the 15 second ceiling of the previous generation. That is enough room for an actual scene: a setup, a beat of dialogue or action, and a payoff, all in one take.

The audio is not an afterthought. Dialogue, ambient sound, and effects are generated together with the picture, so lip movement matches speech and footsteps land when feet hit the ground. For short ads, product demos, and social clips, a single generation can now carry the entire piece without any editing at all.

Multi-Round Extension: From Clips to Short Films

Multi-round extension lets you continue a finished clip with another generation pass, and then another, chaining takes into videos that run for multiple minutes. Characters, visual style, and audio carry across rounds, so the result plays as one continuous piece rather than a stack of disconnected clips.

This is the feature that moves Seedance 2.5 from "clip generator" to "short film tool". You can rough out a full 2 minute narrative by extending scene after scene, reviewing each round before committing to the next.

Flexible Referencing: Up to 50 Inputs Per Generation

Seedance 2.5 accepts up to 50 multimodal references in a single pass: up to 30 images, 10 video clips, and 10 audio clips. References can pin down a character's face and wardrobe, a location, a camera move, a motion style, or a specific voice.

In practice this means you direct with assets instead of adjectives. Feed it three angles of your character, a still of your set, a reference clip with the camera move you want, and a voice sample, and the model composes a scene that respects all of them. Consistency across shots stops being a prompt engineering trick and becomes an input you control.

Timestamp-Level Editing

Timestamp-level editing lets you change a specific moment inside a generated clip without touching the rest. If the line reading at second 14 is wrong, or a hand glitches at second 22, you edit that segment and keep everything around it.

This fixes the most expensive habit in AI video work: regenerating a whole clip, and losing everything that was already good about it, to repair one bad second. Iteration gets faster and cheaper because good takes stay locked while you polish the weak spots.

Green Screen, Camera Perspective, and Clay Render Editing

Seedance 2.5 ships with editing modes that used to require a studio pipeline. Green screen generation produces subjects on clean keyable backgrounds for compositing. Camera perspective editing reframes an existing shot from a different angle. Reference-based editing reshapes a scene, swaps elements, or restyles footage while preserving its structure.

The most unusual addition is clay render referencing: hand the model a rough untextured 3D blockout, and it renders the shot with physically plausible lighting and materials. For anyone doing previz or product visualization, that is a direct bridge from blockout to finished frame.

More Believable Skin, Eyes, and Textures

ByteDance also tuned the model against the telltale AI look. Skin, eyes, hair, and fine surface textures render with noticeably more detail, and faces hold up under close framing instead of going waxy.

Combined with synchronized dialogue, this matters most for people-centric content: talking head clips, UGC-style ads, and character-driven scenes read as footage first and generated content second.

Getting Started Today

Sign up for Runable at runable.com if you haven't already. Seedance 2.5 is included on the Pro plan at $20/month (most popular) and the Max plan at $100/month, as of August 2026. Pick Max for the fastest generation speeds and the heaviest workloads, like multi-minute extension chains.

Then navigate to AI Videos, select Seedance 2.5, and describe your first take. Attach your references (character images, location stills, motion or voice clips), generate your 30 second take, fix any weak moments with timestamp edits, extend the clip if you need a longer piece, and export.

FAQ

What is Seedance 2.5?

Seedance 2.5 is ByteDance's latest AI video model, announced on July 31, 2026 and built around one-take creation and flexible referencing. It generates up to 30 seconds of video with synchronized audio in a single pass, extends clips into multi-minute videos, and accepts up to 50 multimodal references. It is available on Runable from day one.

How do I use Seedance 2.5 on Runable?

Sign up at runable.com on a Pro or Max plan, open the AI Videos section, and select Seedance 2.5. Write a prompt for your take, attach reference images, video clips, or audio if you want tight control, and generate. From there you can fix moments with timestamp edits, extend the clip, and export.

Is Seedance 2.5 free to use?

Seedance 2.5 requires a paid Runable plan: Pro at $20 per month (most popular) or Max at $100 per month, as of August 2026. Runable's free tier still includes other video models, such as Kling 3.0, so you can try AI video at no cost before upgrading for Seedance 2.5.

How long can Seedance 2.5 videos be?

A single Seedance 2.5 generation produces up to 30 seconds of continuous video with synchronized audio, double the previous 15 second limit. With multi-round extension, you can keep continuing a finished clip while characters, style, and sound carry over, which takes a single project to multi-minute lengths.

Can Seedance 2.5 edit existing videos?

Yes. Timestamp-level editing changes a specific moment in a clip without regenerating the rest. Reference-based editing reshapes scenes, swaps elements, or restyles footage while preserving structure. Camera perspective editing reframes shots from new angles, and green screen generation produces clean keyable subjects for compositing into existing footage.

What references does Seedance 2.5 support?

One generation accepts up to 50 multimodal references: up to 30 images, 10 video clips, and 10 audio clips. Use images for characters, wardrobe, and locations, video clips for motion and camera language, and audio for voices. It even accepts rough clay renders from 3D blockouts and lights them physically.