Seedance reference to video

Generate with references

What references do

A reference is guidance, not a starting frame. Reference images tell the model what a subject, style or environment looks like from several angles; the generated shot can then place that subject in any scene the prompt describes. Reference videos contribute motion, pacing and camera language. Reference audio contributes rhythm, and on 2.5 can drive timing for music visuals. ByteDance describes this flexible referencing as the central idea of Seedance 2.5, alongside white-model and green-screen control for driving a scene from a blockout or a keyed subject.

Seedancee is an independent platform; the models are ByteDance's and are served by Kie.ai. The limits below are read from our model registry, which we verify against Kie's documentation, and the generator enforces them before you spend credits.

Reference limits by model

As documented by Kie.ai on 29 September 2026: Seedance 2.5 accepts reference videos up to 200 MB and 30 seconds each with a 30-second combined cap, and audio up to 30 seconds; Seedance 2.0 variants accept videos up to 50 MB and 15 seconds with a 15-second combined cap, and audio up to 15 seconds. Video references should be 480p or 720p, 24–60 fps, MP4 or MOV; audio WAV or MP3 up to 15 MB. Seedance 1.5 Pro does not accept reference media and is omitted from the table.

Preparing references

1. 01

   Characters

   Three to six images: front, three-quarter, profile, full body, plus one close-up. Plain backgrounds, consistent clothing. Describe the same features in the prompt so text and image agree.

2. 02

   Products

   Two to four images on a neutral background at the angles you want in the shot. Include one of any label or texture that must survive.

3. 03

   Motion and camera

   A short clip of the move you want (a whip pan, a slow orbit, a dance step). Trim it to the relevant seconds; the combined video duration is capped.

4. 04

   Audio

   A licensed WAV or MP3 trimmed to the section you need. Only upload audio you have rights to; the content policy applies to inputs as well as outputs.

5. 05

   Own the rights

   References of real people, copyrighted characters or third-party brands are not allowed unless you hold the rights and consent. The moderation pipeline checks inputs.

Prompting with references

Refer to the references explicitly ("the character from the reference images", "the camera move from the reference clip") and still describe the scene in full. The prompt sets the new scene; the references keep the identity and motion stable. When mixing several references, say which is which: "Use the referenced product for the bottle and the referenced clip for the camera move."

Example reference-to-video prompts

The character from the reference images, a woman with silver braids and a patched brown flight jacket, walks through a crowded night market, stops at a lantern stall and holds a paper lantern up to her face, smiling. Medium tracking shot, warm practical light, shallow focus. Sound: crowd murmur, paper crinkling, a distant drum.

Use the referenced product images for the trail-running shoe and the referenced clip for the camera move: a fast low orbit that settles into a slow push-in. Shoe on wet slate at dawn, mist, cold blue light with a warm rim. No text. Sound: wind, a single drip.

Abstract visualizer synced to the referenced audio: liquid chrome ribbons twist through black space and pulse with the beat, magenta and cyan reflections, seamless motion, no text, no people.

In the painted style of the reference images, a small fishing boat returns to a harbour at sunset, gulls circling, nets piled on the dock. Slow pan right. Sound: gulls, water lapping, a bell.

Seedance 2.5 vs 2.0 for references

Seedance 2.5 is the better reference model on almost every axis: 30 reference images instead of 9, longer video and audio references, 30-second output and the ability to extend the result. Seedance 2.0 is the choice only when the final needs 4K, which 2.5 does not offer. Seedance 2.0 Fast and Mini share 2.0's reference limits at 480p/720p and are useful for testing a reference set cheaply before committing.

Cost

Reference images and audio carry no surcharge. Including at least one reference video adds a flat surcharge on top of the duration, resolution and audio cost, visible in the estimator as soon as you tick the reference-video option. Failed generations are refunded automatically. Reference-to-video with images works on both plans (Starter covers Seedance 2.0 Mini and 2.0 Fast at up to 720p and 10 seconds); reference videos as inputs, and the 2.5 and 2.0 models, need the Pro plan.