Seedance 2.0 omni reference workflow

Seedance 2.0 Omni Reference: What It Is and How It Works

This Studio combines text prompts with image and video references. Use images for appearance, text for intent and video for motion. Audio-file references are not supported; generated sound is a separate setting.

Direct answer

Seedance 2.0 omni reference works by separating creative control across multiple inputs. Use an image to preserve the subject, a prompt to define the action and a video clip to guide movement. Add precise music or voiceover later in an editor.

Best input order

Start with the image reference, then add text, then add a motion reference only if needed. Too many controls at once can create conflicts. For prompt wording, pair this workflow with the plain-English prompt guide.

Reference stack

  • Hero image: lock the subject, color palette, product shape, or character outfit.
  • Text prompt: explain the exact action and what may change from the reference.
  • Motion clip: guide body movement, camera path, object rotation, or pacing.
  • Generated sound: enable Audio to request sound for the clip. This does not upload a soundtrack or guarantee beat synchronization.

A practical workflow

  1. Upload the clearest reference image you have. Remove text overlays, watermarks, and clutter before using it.
  2. Write a short text prompt that states what should move and what should stay unchanged.
  3. Run a short draft. If the main motion works, test generated sound or a longer duration while keeping the other settings fixed.
  4. For repeated scenes, save a reference set and move successful shots into the API batch workflow.

Prompt examples for references

Use the uploaded product image as the identity reference.
Preserve the exact bottle shape, label color, and cap.
Animate only the environment: cold mist rolls behind the bottle, droplets form on glass.
Camera: locked-off macro shot, 6 seconds, no new text.

When not to add more references

If a draft is almost correct, do not add another reference by default. First tighten the text: "same face", "same logo", "fixed camera", "front view only", or "no scene change." Add a motion clip only when the movement itself is hard to describe. For exact music beats or voiceover timing, finish the clip in an editor.

Build a clean reference stack

Begin with one strong image and one direct prompt. Add a video reference only when it solves a visible control problem.

Start creating

Continue the workflow

Apply the same prompt, reference and budget checks when moving between model versions.

Return to the complete guide library or compare API credit costs.