Try Fable AI · ByteDance video model
Seedance 2.5 is ByteDance’s next-generation AI video generator. It produces a single continuous 30-second shot in one pass, accepts up to 50 multimodal references, and generates synchronized audio and video together. Run text-to-video and image-to-video right here, then export cinematic clips at up to 4K.
Seedance 2.5 is the latest video generation model from ByteDance, introduced at the Volcano Engine FORCE conference in June 2026. ByteDance skipped straight from Seedance 2.0 to 2.5, framing the release as a generational jump rather than an incremental update — and the headline numbers reflect that: a native 30-second single-shot clip, up to 50 multimodal reference inputs, and a unified audio-video architecture.
The most practical change is duration. Earlier video models topped out around 15 seconds before quality drifted, forcing creators to stitch shorter segments together. Seedance 2.5 renders a full 30-second shot in a single pass, holding character appearance, lighting, and motion style steady across the whole clip. That makes it suitable for complete scenes, narrative beats, and product spots that previously required manual editing.
Seedance 2.5 also changes how much direction you can give the model. Instead of a handful of reference images, it accepts up to 50 multimodal inputs — images, video, audio, style references, and 3D blockouts — so you can lock a character, a look, a camera path, and a soundscape in one generation. Use the generator above to try text-to-video and image-to-video, then refine with localized region editing.
Seedance 2.5 focuses on longer, more controllable, audio-synced video. These are the capabilities that set it apart from earlier AI video models.
Generate a continuous 30-second clip in one pass — no stitching and no visible seams. The model maintains subject consistency, lighting, and motion style across the full duration, roughly doubling the previous single-clip limit.
Guide a generation with up to 50 reference inputs at once, including images, video clips, audio, style references, and 3D blockouts. That gives you far more control over character identity, composition, and motion than a text prompt alone.
Visual and audio signals are produced together in the same pass rather than added afterward, so dialogue, footsteps, impacts, and ambience line up natively with the action on screen.
Redraw a single element — a character’s outfit, a background object, a product in the scene — without regenerating the entire clip, keeping the rest of the frame consistent.
Feed a low-fidelity 3D blockout to pre-stage camera moves and composition, giving directors frame-level control before committing to a full-quality render.
Render directly at up to native 4K (3840×2160) with 10-bit color for smoother gradients and more grading headroom, with noticeably tighter prompt following so you reach a usable result in fewer tries.
Seedance 2.5 works from a written brief, a still image, or a set of references — pick the starting point that matches your workflow.
Describe the subject, action, setting, camera, and mood, and Seedance 2.5 turns that brief into a finished moving shot with synchronized sound.
Start from a still image and animate it into a moving scene, or set a first and last frame to guide how the shot begins and ends.
Combine images, clips, audio, and style references to lock a consistent character, look, and motion across multiple generations.
Both models are available here. Seedance 2.5 extends duration, reference capacity, and editing control; Seedance 2.0 remains a fast, capable option for shorter synced clips.
| Feature | Seedance 2.5 | Seedance 2.0 |
|---|---|---|
| Max single-clip duration | Up to 30s, single shot | Up to 15s, multi-shot |
| Reference inputs | Up to 50 multimodal | Up to 12 (9 images, 3 video, 3 audio) |
| Resolution | Up to native 4K, 10-bit | Up to native 4K, 10-bit |
| Audio | Unified joint audio-video | Joint audio-video |
| Localized region editing | Yes | Not available |
| 3D blockout input | Yes | Not available |
| Prompt adherence | ~20% higher | Baseline |
| Best for | Long, complex, multi-reference shots | Short synced clips and fast iteration |
Longer single shots and richer reference control open up work that previously needed an editing timeline.
Produce complete 30-second scenes with consistent characters and camera language for trailers, concept films, and social storytelling.
Generate product spots and lifestyle scenes from references, then swap a product or background with localized editing instead of re-rolling the whole clip.
Use joint audio-video generation to align motion, beats, and ambience for music videos, lyric pieces, and performance clips.
Stage shots with 3D blockouts and reference frames to previsualize camera moves and composition before a full production render.
Lead with the subject and action, then add setting, camera, lighting, and mood. Concrete direction outperforms broad quality adjectives.
Attach images, frames, clips, audio, or style references to lock identity, look, and motion. Keep each reference relevant to the shot you want.
Choose the clip length, aspect ratio, and resolution that match your final placement, and enable audio when you need synchronized sound.
Generate, review the result in your history, and use localized region editing to adjust a single element without regenerating the whole clip.
Seedance 2.5 is ByteDance’s next-generation AI video generation model. It creates a native 30-second single-shot clip in one pass, accepts up to 50 multimodal references, and generates synchronized audio and video together, with support for text-to-video and image-to-video.
Seedance 2.5 generates a continuous shot of up to 30 seconds without stitching segments together, roughly double the typical 15-second limit of earlier models. Shorter durations are also available for quicker iteration.
Seedance 2.5 accepts up to 50 multimodal reference inputs in a single generation, including images, video clips, audio, style references, and 3D blockouts — a large increase over the 12 references supported by Seedance 2.0.
Yes. Seedance 2.5 uses a unified audio-video architecture that produces visuals and sound together in the same pass, so dialogue, effects, and ambience stay synchronized with on-screen action instead of being added in a separate step.
Yes. You can animate a still image into a moving scene, or set a first and last frame to control how the shot begins and ends. Seedance 2.5 also supports text-to-video and reference-driven generation.
Seedance 2.5 can render at up to native 4K (3840×2160) with 10-bit color for smoother gradients and more color-grading headroom. Lower resolutions are available for faster drafts.
Seedance 2.5 extends single-clip duration to 30 seconds, raises reference capacity to 50 inputs, and adds localized region editing and 3D blockout staging. Seedance 2.0 generates synced clips up to 15 seconds with up to 12 references and remains a fast option for shorter videos.
Seedance 2.5 runs on the same credit system as the other models on Try Fable AI. The exact credit cost depends on the duration and resolution you select, and is shown in the generator before you start a job.
Enter a prompt, add your references, and generate a cinematic clip with synchronized audio in a single pass.
