Loading page…
No preview yet
Adjust the settings on the left and generate a video to get started.
Seedance 2.5 is live for text-to-video, image-to-video and multimodal reference generation at 480p, 720p or 1080p.
ByteDance's Seedance 2.5 generates 4 to 30 seconds of synchronized audio and video from text, frames or up to 50 multimodal references. It is live in the generator above through EvoLink.
Seedance 2.5 raises the ceiling on every part of the pipeline: longer single-pass output, far more references, and editing you can aim at a specific moment.
Seedance 2.5 generates a full 30-second clip in one pass, up from 15 seconds. Multi-round extension carries that out to several minutes while holding characters, environment and narrative pacing steady.
Audio and visuals come out of one unified multimodal architecture, so sound and picture are aligned from the first frame. Seedance 2.5 keeps that alignment across extension rounds.
One pass accepts up to 30 reference images, 10 video clips and 10 audio clips. That is the 50-reference figure, broken down — video and audio are first-class references, not just style hints.
Point Seedance 2.5 at a time range and change the narrative, camera, movement or rhythm inside it. Characters, actions and plot in a specific clip can be revised after the video exists.
Feed in a textureless 3D model to lock spatial structure, character poses, motion paths and camera angles. Seedance 2.5 reads the blockout and derives light direction, colour temperature, intensity and shadow from it.
Replace a background while the subject stays intact, with clothing, hair, gait and lighting responding physically. Camera movement can be re-aimed without disturbing the performance underneath.
Seedance 2.5 keeps 2.0's joint audio-video architecture and pushes length, references and editing control well past it. The Seedance 2.0 column is what the generator on this page does right now.
One continuous 30-second take replaces a stitched sequence of short shots.
Extension rounds carry characters, environment and pacing forward, so a multi-minute video holds together.
Same unified architecture in both. Seedance 2.5 adds sync that survives extension rounds.
Video and audio become reference material in their own right, not just an image board.
Editing moves from re-rolling a prompt to aiming at a moment, a subject or a camera path.
Camera position and staging can be locked before generation instead of coaxed out of a prompt.
EvoLink's live Seedance 2.5 route supports 480p, 720p and 1080p output. Seedance 2.0 Standard remains available for 4K workflows.
Both generations are live in the model picker, with Seedance 2.5 selected by default on this page.
Real Seedance 2.5 output posted publicly on X, embedded with attribution intact.
Choose Seedance 2.5 in the generator and build text, frame-guided or multimodal reference videos directly.
Choose Seedance 2.5, then select text-to-video, image-to-video or reference-to-video for the assets you have.
Name the subject, the action, the camera move and the light in one line. Add reference images to hold a character or a look steady across takes.
Pick 4 to 30 seconds, 480p to 1080p, an aspect ratio from 21:9 through 9:16, and leave audio generation on.
EvoLink's Seedance 2.5 routes are live. This page now sends text, frame and multimodal reference jobs directly to Seedance 2.5.
The official Seed launch post documents 30-second single-pass audio-video, the 30 + 10 + 10 reference breakdown, timestamp editing and clay-render blockout.
The integration uses EvoLink's unified video endpoint with dedicated Seedance 2.5 text, image and reference model IDs.
Seedance 2.0, Fast and Mini run in the generator above with text, image and reference input, 4 to 15 second clips, up to 4K on the standard tier, and audio generation on by default.
August 20, 2026
What Seedance 2.5 does, what changed since 2.0, and how to run it here.
Seedance 2.5 is ByteDance's video generation model. It produces a 30-second clip with synchronized audio in a single pass, accepts up to 30 reference images, 10 video clips and 10 audio clips, and supports timestamp-level editing of narrative, camera, movement and rhythm.
Single-pass length doubles from 15 to 30 seconds and extends to several minutes across rounds. References grow to 30 images plus 10 video and 10 audio clips. Editing gains timestamp targeting, green-screen replacement, camera-perspective changes and clay-render 3D blockout.
Thirty seconds in one pass. Longer pieces are built with multi-round extension, which carries characters, environment and narrative pacing forward so a multi-minute video stays consistent shot to shot.
Yes. Seedance 2.5 uses a unified multimodal architecture that generates audio and video jointly, so sound and picture are aligned from the start rather than dubbed on afterwards — and stay aligned through extension rounds.
A textureless 3D model you supply to define spatial structure, character poses, motion paths and camera angles. Seedance 2.5 reads that blockout and renders realistic light direction, colour temperature, intensity and shadow on top of it.
EvoLink's current Seedance 2.5 API routes support 480p, 720p and 1080p, not 4K. Use 1080p for final output or select Seedance 2.0 Standard for an available 4K route.
Yes. Seedance 2.5 is selected by default in the generator above and supports text, first/last-frame and multimodal reference workflows through EvoLink.
Generate 4 to 30 second videos from text, first and last frames, or multimodal references with synchronized audio.
More than resolution. Native 4K with 10-bit color delivers richer color gradation, smoother transitions, and greater flexibility for professional workflows.
i dont think people realize how big this is Seedance 2.5 now can generate 30s 4K videos from one prompt.. with up to 50 ref.. one click filmmaking is here
SeeDance 2.5 is crazy!
