Seedance 2.5

Seedance 2.5: AI Video Generation Model (ByteDance)

What is Seedance 2.5?

Seedance 2.5 is ByteDance's next-generation AI video generator from the Seed model family. It turns a text prompt, a starting image, or a stack of multimodal references into a coherent, cinematic clip with synchronized native audio. Announced on June 23, 2026 at the Volcano Engine FORCE conference and launched publicly on July 31, 2026, it succeeds the top-ranked Seedance 2.0 line and targets longer, more controllable single-take video. On Segmind it runs as a synchronous text-to-video and image-to-video API: send a prompt, optionally add a first frame or references, and receive an MP4 with an audio track. Durations run 4 to 30 seconds at 480p or 720p.

Key Features

• •Native audio: dialogue, sound effects, and music generated jointly with the visuals, so motion and sound stay in sync.
• •Multi-shot storytelling in one prompt using Shot 1: / Shot 2: scripting for continuous scene changes.
• •Deep reference control: up to 30 reference images, 10 reference videos, and 10 reference audios, cited inline as @Image 1, @Video 1, @Audio 1.
• •Modes: text-to-video, image-to-video (first_frame_url), first-to-last-frame transitions (last_frame_url), and reference-to-video.
• •Directable camera language, seed control, and a bitrate_mode for higher fidelity.

Best Use Cases

Seedance 2.5 fits ad spots, product demos, social shorts, music-driven clips, and pre-visualization. The reference stack keeps a character, product, or palette consistent across a shot. In testing, a generated landscape animated cleanly from a single first frame into a 5-second 720p clip with realistic water and mist motion and coherent ambient audio, with no visible artifacts.

Prompt Tips and Output Quality

Write the shot as evolving motion: subject and action first, then scene, style, camera move, and audio. Name what each reference controls. When you pass a first_frame_url, set aspect_ratio to adaptive so output follows the source frame. Avoid real human faces in first frames, which policy blocks; use reference images for characters instead.

FAQs

Does Seedance 2.5 generate audio? Yes. Enable generate_audio to co-produce synchronized sound in the same pass.

How long can videos be? On Segmind, 4 to 30 seconds per generation at 480p or 720p.

Can I do image-to-video? Yes, pass a first_frame_url; add last_frame_url for a guided transition.

How many references can I use? Up to 30 images, 10 videos, and 10 audios per request.