Back to Home

Seedance 2.5 AI Video Generator & Editor

Seedance 2.5 is ByteDance Seed's joint audio-video generation model for up to 30-second storytelling, multimodal reference and directed editing. Use it for generation, editing and extension in the Flux Art multi-model AI visual creation and production platform.

Create with Seedance 2.5
Explore six video workflows
Multimodal video direction

Do more than generate a clip—assign every input a job

Seedance 2.5 provides six starting points in Flux Art. Decide whether visual facts come from text, boundary frames or references, then focus the prompt on action, camera, sound and protected details.

  • Long-form beat

    Let one action unfold on a complete timeline

    Plan opening, development and ending across up to 30 seconds before adding camera and sound; do not treat duration as room for unrelated events.

  • Reference direction

    Give each input one clear responsibility

    Let images own appearance, video own motion and audio own rhythm, then explain how they meet in the prompt.

  • Edit and extend

    Protect first, then describe change

    Lock identity, structure, composition, camera and sound in the source before narrowing the edit or continuation to a reviewable change.

Six starting points

Choose the workflow before the prompt

Six routes determine how media enters the task and which ratio, duration and required inputs apply.

  • T2V

    Text to video

    Start from one complete brief for scene, subject, action, camera and sound.

  • FIRST

    First-frame video

    Let the first frame provide visual facts while the prompt adds change over time.

  • A → B

    Boundary-frame video

    Use two boundary images for the start and result while the prompt owns the path.

  • REF

    Multimodal reference

    Combine image, video and audio with explicit identity, object, motion and rhythm jobs.

  • EDIT

    Video editing

    Start from existing footage and rewrite one explicit, reviewable layer.

  • EXTEND

    Video extension

    Inherit the source ending and continue motion, camera, space and sound.

Reference stack

Up to 50 materials, not 50 ambiguous instructions

Flux Art validates images, video and audio separately. The total is a capacity limit; clear responsibilities, consistent indexes and non-conflicting inputs matter more.

50total material limit
  • 30images

    Own visual facts

    Identity, product structure, place, wardrobe, material and boundary states.

  • 10videos

    Own temporal reference

    Motion paths, performance rhythm, camera movement, or source footage for editing and extension.

  • 10audio files

    Own sound and rhythm

    Music rhythm, sound atmosphere and audiovisual cues; reference mode also accepts audio-only input.

Production controls

Where Seedance 2.5 fits

It is most useful when a shot has several input constraints, needs continuity, or requires a directed change to existing footage.

  • 01

    Shots up to 30 seconds

    Use 4–30 seconds for a complete action, product demonstration or short narrative beat with fewer stitched generations.

  • 02

    Multimodal reference direction

    Assign people, products, places, motion, rhythm or sound across image, video and audio inputs, with up to 50 total materials.

  • 03

    Directed existing-video edits

    Name the object, wardrobe, environment or detail allowed to change while protecting composition, subject, motion and timeline.

  • 04

    Video extension and continuation

    Continue from a source clip's final state so new motion and camera direction inherit space, direction, rhythm and sound.

Director briefs

Seedance 2.5 director briefs

Define each input's job before writing temporal change and protected boundaries; avoid giving several inputs the same responsibility.

  • Multimodal reference

    Multi-input product launch

    Image 1 owns the headphone shape and logo, image 2 owns the deep-blue plinth material, video 1 owns only the low orbit into a front view, and audio 1 owns rhythm. Ten seconds: 0–3s reveal with a narrow light, 3–8s orbit to the beat, 8–10s settle on the front hero. Preserve proportions, controls and marks; add no text.

  • Existing-video edit

    Wardrobe-only video edit

    Preserve the source subject's identity, face, hair, walking speed, camera motion, street composition and ambience. Change only the black jacket to a clean cobalt trench that follows the original movement; do not alter hands, pedestrians, light direction or timeline.

  • Video extension

    Extend a rainy arrival shot

    Continue eight seconds from the source clip's final frame. The subject pauses at the tram door, boards, and the shoulder-height camera follows. Carry forward rain density, teal street light, wet wardrobe and footsteps. End outside the window as the door closes; add no new people or text.

  • Text-to-video

    30-second one-take brand story

    Thirty-second continuous take. Begin outside a morning studio and slowly follow a designer through final assembly, packaging and delivery of [product]. Use foreground occlusion to connect three spaces while preserving wardrobe, product structure and warm morning light. Transition ambience from tools to paper and city sound, ending on the customer receiving the product; add no captions.

  • Start-and-end-frame video

    Boundary-frame location transition

    Start from [subject or product] in the first frame and arrive at [destination] in the final frame. Over 12 seconds, move the camera along [clear path] and use [smoke, doorway or fast occlusion] for a natural location transition. Preserve identity, proportions, colors and light direction, landing precisely on the final-frame composition with no added text.

  • Audio reference

    Audio-led rhythm sequence

    Audio 1 owns only rhythm, pauses and energy changes; image 1 owns subject appearance. Structure 15 seconds as three connected actions: a slow reveal for five seconds, faster camera movement on the main beat for six, then a stable four-second finish as the audio resolves. Preserve appearance and scene direction; infer no visual facts from the audio.

  • Multimodal reference

    Multi-character continuity scene

    Images 1, 2 and 3 own the identities and wardrobes of three people; image 4 owns the café. In one 20-second shot, A sits, B passes a document, and C enters from the background and stops beside the table. Preserve faces, wardrobe, seating and eyelines; keep dialogue short and non-overlapping, with one slow camera orbit.

  • Existing-video edit

    Background-only video edit

    Preserve the source person, action, timeline, camera, foreground props and audio. Replace only the plain background with [target environment]. Match light direction, contact shadows, wardrobe wind response and ground reflections; do not alter the silhouette, speed or crop.

  • Existing-video edit

    Camera-only video edit

    Keep the source subject, action, performance, setting, lighting, audio and duration unchanged. Change only the camera: 0–4s slow low-angle push, 4–8s track right while centering the subject, 8–12s rise to a slight overhead view and settle. Add no cuts and do not change action timing.

  • Product video

    Continuous product demonstration

    Create a continuous 16-second demonstration for [product]. Show the full design at 0–4s, demonstrate [core feature] with hands at 4–10s, reveal the real use result at 10–14s, then return to a stable product close-up at 14–16s. Preserve dimensions, controls, ports and branding; keep hands natural and add no copy.

Create and review

How to use Seedance 2.5 in Flux Art

Capabilities reviewed against ByteDance Seed's official model page, launch article and the current Flux Art integration on August 26, 2026. Available controls and credit costs follow the live workspace.

Create with Seedance 2.5
  1. Choose the input workflowStart from text, a first frame, boundary frames, multimodal references, video editing or extension; the route decides which facts come from media.
  2. Assign media responsibilitiesGive every input a clear job such as identity, product structure, camera motion, rhythm or ambience, then reference it by its prompt index.
  3. Write change over time and reviewWrite opening state, main action, camera and sound cues, and ending in order; then review identity, structure, motion, continuity and audiovisual alignment separately.
Source notes

Official sources and page information

Seedance 2.5 capabilities are summarized from official provider materials. Available modes, parameters and pricing follow the current Flux Art workspace.

Published

Compare related AI video models

Seedance 2.5 FAQ

What is Seedance 2.5?

Seedance 2.5 is ByteDance Seed's joint audio-video generation model, officially launched on July 31, 2026, for up to 30-second storytelling, multimodal reference and directed editing. Flux Art currently connects it as a 4–30 second workflow with 480p and 720p output.

Which Seedance 2.5 workflows are available in Flux Art?

The current workspace provides six routes: text-to-video, first-frame video, start-and-end-frame video, multimodal reference, existing-video editing and video extension. Select the route before adding media because each route validates different inputs.

How do I use Seedance 2.5 in Flux Art?

Use the generator on this page, choose one of the six workflows, add only the media required by that route, describe the action, camera, sound and protected details, select duration and resolution, then review the live credit estimate before submitting.

What duration, resolution and aspect ratios does Seedance 2.5 support?

Flux Art currently supports 4–30 second output at 480p or 720p. Text-to-video offers fixed ratios including 16:9, 4:3, 1:1, 3:4, 9:16 and 21:9; media-led workflows use adaptive ratio when the source determines the composition.

How many reference materials can I use?

The current multimodal-reference limit is 50 materials in total: up to 30 images, 10 videos and 10 audio files. Reference videos and reference audio are each limited to 30 seconds of total input duration, and the live uploader applies the current file-size and format checks.

Can I use audio without an image or video reference?

Yes. The current multimodal-reference workflow accepts audio-only reference input. Explain whether the audio should control rhythm, ambience, dialogue timing or another specific responsibility so it does not conflict with the written direction.

What is the difference between video editing and video extension?

Editing starts from an existing clip and changes a defined layer while protecting the source timeline, subject, composition or camera. Extension begins from the source ending and continues the action, camera, space and sound into a new segment.

Why does the aspect ratio switch to Adaptive for media-led workflows?

A first frame, boundary frames or source video already establishes the composition. Adaptive lets the model follow that media instead of forcing an unrelated fixed canvas; Flux Art disables the ratio control in those routes to prevent contradictory input.

How are points calculated for Seedance 2.5?

Flux Art prices Seedance 2.5 from the selected resolution and output duration. When a task contains reference video, the measured total reference-video duration also participates in the calculation. The workspace shows the current estimate before submission and rounds the final task charge to the configured point unit.

Can Seedance 2.5 generate sound or use web search?

Yes. Flux Art currently exposes audio generation and optional web search for Seedance 2.5. Use only the controls needed for the shot, and review dialogue, ambience, music and timing separately before publishing.

How should I write a prompt with several reference files?

Give every file one job and refer to it consistently: an image can own identity or product shape, a video can own motion or camera, and audio can own rhythm. Then write the opening, change, camera and ending in time order and list details that must remain unchanged.

What are the main Seedance 2.5 limitations?

ByteDance Seed notes that complex-motion physics and interactions between multiple subjects can still lose plausibility or stability. Review anatomy, object contact, subject identity, timing, dialogue and continuity after every generation, especially in long or crowded scenes.

How is Seedance 2.5 different from Seedance 2.0?

In Flux Art, Seedance 2.5 is the broader production workflow for output up to 30 seconds, larger multimodal reference sets, targeted video editing and extension. Seedance 2.0 remains a useful alternative for product videos, story shots and shorter generation workflows.

Can I try Seedance 2.5 for free in Flux Art?

Flux Art provides sign-up credits for an initial trial. The generator displays the current estimated credit cost before submission; available credits, rates and plan terms can change, so use the live workspace and pricing page as the source of truth.

Can Seedance 2.5 videos be used commercially?

Commercial use depends on the current Flux Art and model-provider terms and on your rights to prompts, people, brands, music and reference media. Review the latest terms and secure the necessary permissions before publishing client, advertising or branded work.