BytePlusSeedance 2.5

MULTIMODAL VIDEO GENERATION / 2026

Turn imagination
into production-ready video.

Long-form storytelling × high-reference fidelity × precise editing × multilingual creation

Reference up to 50 images, videos, and audio files. Generate up to 30 seconds, edit and extend at professional quality, and tell stories in more than 10 languages—all within one production flow.

Explore the demos See the workflow
30SEC
LONG FORM
50ASSETS
REFERENCE
10+LANGUAGES
SUPPORTED

HIGH-FIDELITY PREVIS → FINAL OUTPUT SEEDANCE 2.5

01OVERVIEW

Multimodal becomes a production workflow.

Move beyond one impressive generation to a repeatable, scalable process that keeps every video ready to deliver.

ONE PRODUCTION FLOW

Reference → Generate → Edit → Extend

Connect source assets and creative intent in one controlled flow, from first prototype to final delivery.

30sMaximum duration

Tell a longer story in one generation.

50Reference assets

Combine images, video, and audio.

10+Languages

Create for global audiences.

02CAPABILITIES

One model, from reference to editing and extension.

01REFERENCE

From reference to intent, faithfully.

Carry subjects, motion, style, audio, storyboards, and keyframes into a new production with high fidelity to both the source assets and the intended composition.

  • Subject
  • Motion
  • Style
  • Audio
  • Storyboard
  • Keyframes
02EDIT

Change only what needs changing.

Edit subjects, wardrobe, backgrounds, expressions, cameras, and audio—even within a defined time range—while preserving everything else.

  • Add & remove
  • Reference edit
  • Time range
  • Audio edit
03EXTEND

Continue without breaking continuity.

Extend a story forward or backward while preserving visual and audio consistency, or build seamless transitions between separate clips.

  • Extend before
  • Extend after
  • Audio continuity
  • Transitions
04ONE-CLICK

From assets to a finished cut in one step.

Turn multiple images and videos into a short-form edit complete with text, stickers, transitions, and a social-ready format.

  • Multiple assets
  • Auto edit
  • Social formats
  • Short-form video
2.1A / WHITE MODEL REFERENCE

Keep the motion and camera design.
Render the final cinematic world.

Preserve the structure and direction of a white-model previs, then render it as a synchronized final scene.

INPUT / PREVISOUTPUT / FINAL
2.2STORYBOARD

From a nine-frame plan to a 30-second story.

Combine storyboard composition, movement, and narrative with dedicated environment and character references.

Nine-frame storyboard
Environment referenceJapanese-labeled robot character referenceCharacter reference
01 / STORYBOARD03 / REFERENCES
SEEDANCE 2.5 OUTPUT00:30
03WORKFLOW

Provide the assets. State the intent. Produce video you can use.

  1. 01

    Map every asset

    Number each image, video, and audio file, then state exactly what the model should reference.

  2. 02

    Define the story in one sentence

    Subject + location + event + genre or style + camera direction.

  3. 03

    Design the timeline

    Use whole-second timestamps to direct visuals, camera moves, action, dialogue, and sound.

  4. 04

    Lock the global requirements

    Specify the framing, light, sound, mood, and quality that must remain consistent throughout.

  5. 05

    Refine only what matters

    Edit or extend a defined range until the result is ready to deliver.

04DEMOS

Edit only what you want to change—in words.

Change a subject or a defined time range while preserving the composition, camera, lighting, and timing.

BEFOREIn her 20s · tearful
AFTERIn her 60s · tears to smile
EDIT INSTRUCTION

“Keep the composition, camera, lighting, and timing. Change the woman from her 20s to her 60s. Her tears gradually become a smile.”

MORE PRODUCTION MODES

Any asset, format, or editing objective.

Open the complete guide (Japanese) ↗
01 / TRANSITION

Two different clips. One continuous camera experience.

Infer the missing movement between two uploaded videos and generate a natural bridge.

02 / ONE-CLICK

Choose the assets. Get the short.

Turn multiple inputs into a social-ready format.

03 / REFERENCE EDIT

Keep the motion. Replace the world.

Control the location, wardrobe, and subject with reference images.

04 / AUDIO EDIT

Change the language. Keep the performance.

Synchronize English dialogue with natural lip movement and no burned-in subtitles.

PROMPT RECIPE

Write a strong prompt
like a shooting brief.

Replace abstract wishes with explicit asset roles, visible actions, timing, and the elements that must not change.

PROMPT / SEEDANCE 2.5
  1. 01Map references

    Use the person in Image 1, the camera move in Video 2, and the melody in Audio 3.

  2. 02One-line summary

    Subject + location + event + style + distinctive camera movement.

  3. 03Scene progression

    [0–3 sec] Direct visuals, camera, action, dialogue, and SFX at whole-second timestamps.

  4. 04Global requirements

    Light, framing, sound, mood, continuity, and quality. “No subtitles.” “No BGM.”

REFERENCE → SUMMARY → TIMELINE → GLOBAL NOTES
INPUT GUIDELINES

Give every one of the 50 assets a role.

Stable production comes from defining what each input should control—not simply adding more files.

30IMAGES

Images

Up to 4K each

10VIDEOS

Videos

Up to 30 seconds total

10AUDIOS

Audio files

Up to 30 seconds total

50TOTAL

Total assets

Per generation

RECOMMENDED RANGE

Subjects: 1–8 images / 1–5 audio or video files

Storyboards: Up to 15 frames, with minimal text

Video editing: Up to 20 seconds

FROM GENERATION TO PRODUCTION

Move beyond one-off generations.
Build a repeatable video workflow.

Explore the Seedance 2.5 Multimodal Video Generation User Guide for task selection, asset preparation, prompt design, and practical examples.

Open the user guide (Japanese) ↗Back to top ↑