Choose a category
3 models available
4 models available
3 models available
2 models available
4 models available
8 models available
Seedance 2.5 Reference to Video
Cast up to 30 images and say which one plays what.
Creative Vision
Say what each upload is for: “@Image1 for the character, @Image2 for the set.”
Video Parameters
Priced per second — 30 seconds costs 7.5× what 4 does.
720p costs 2.25× what 480p does. Draft at 480p.
Adaptive follows your reference material; pick a ratio to override it.
AI audio synced to the video. Does not change the price.
Cinema Preview
Generation History0 records
Seedance 2.5 Reference to Video
Upload as many as 30 reference images, then tell Seedance 2.5 what each one is: "@Image1 for the character, @Image2 and @Image3 for the set." It casts them into a new shot of up to 30 seconds with audio, instead of averaging them into one vague look. This is the mode that makes 2.5 different from every other model here.
The @ImageN tags are the whole trick — uploads without them are just suggestions
Why Choose Reference to Video
Image to video animates the picture you gave it. Reference to video builds a new shot out of the material you gave it — your character, in your location, doing something neither image shows.
Up to 30 References, Each With a Role
Uploads are numbered in the order you add them and you address them in the prompt as @Image1 through @Image30. The model keeps each one where you put it rather than blending them into a single reference — one image for the lead, another for the co-star, three more for the location.
Consistency Across Shots
Reuse the same character reference in run after run and the face, outfit and proportions carry over. That is what makes a series of clips look like one production instead of five unrelated generations.
Up to 30 Seconds
4 to 30 seconds in one pass, priced per second. Long enough to let a referenced character actually do something — walk in, speak a line, react — rather than pose for five seconds.
Audio Included, Free
Sound is generated with the picture in the same pass and does not change the price. Put dialogue in quotation marks in the prompt and the referenced character speaks it.
Reference Video and Audio Too
Images are not the only material you can hand over. Under Advanced, a clip of up to 30 seconds lends its motion — and lowers the price by about 40% — while an audio track of up to 30 seconds drives the performance.
Failed Runs Are Refunded
Credits are held at submission and returned automatically if the generation fails or the provider never answers. A clip that never arrives costs you nothing.
Direct, Don't Describe
Anything you can show the model is something you no longer have to describe in words — and showing works far better than adjectives for faces, products and specific places.
How to Use Reference to Video
This mode rewards being explicit about roles. The prompt is not just a description here — it is what binds each upload to a job in the shot.
Upload 1 to 30 References
JPEG, PNG or WebP up to 30MB each, and you can add up to 30 of them. Characters, products, locations, style plates. Add them in a deliberate order, because the order is what the numbers in your prompt refer to.
Assign Each One a Role
"@Image1 wields the spear, @Image2 and @Image3 are the courtyard at dusk." Name every upload you actually care about; anything you leave untagged is treated as loose inspiration rather than a specific instruction.
Then Describe the Action
After the casting, write the shot itself: what happens, how the camera moves, what it sounds like. This half is what fills 30 seconds — the references only decide what things look like.
Draft at 480p, Deliver at 720p
30 credits for 4 seconds at 480p is the setting for checking whether the model understood your casting. Once it has, re-run the same references longer and at 720p.
Reference Images Are Not Frames
None of your uploads becomes the first frame — they are material the model draws on. If you want a specific opening image, use Image to Video; for a specific opening and ending, use First to Last Frame.
Seedance 2.5 Reference to Video
FAQ
The @Image syntax, how many references you can use, pricing, and how this differs from image to video.
Uploads are numbered in the order you add them, and you refer to them in the prompt as @Image1, @Image2 and so on. Writing "@Image1 for the swordsman, @Image2 for the temple" tells the model which upload governs which part of the shot. Without the tags it still sees the images, but it decides for itself what they are for.
Up to 30. Seedance 2.5 accepts 50 multimodal references in total — 30 images, 10 videos and 10 audio clips — and this page opens up the image side of that. In practice a handful of well-chosen references beats a pile of similar ones, but the room is there when a scene genuinely needs a whole cast.
Image to video animates the picture you uploaded — your image is the first frame. Reference to video generates a new shot that uses your images as material: the character from one, the setting from another, doing something neither picture shows. Same model, same price, different job.
Between 30 and 430 credits, set by resolution and length: 30 credits for 4 seconds at 480p, 75 for 5 seconds at 720p, and 430 for a full 30 seconds at 720p. The number of reference images does not change the price. The exact cost is shown before you submit.
Close, not identical. Reference images hold identity far better than a written description, and reusing the same reference across runs is the reliable way to keep a series consistent — but this is generation, not compositing, so expect variation in lighting, angle and expression.
JPEG, PNG and WebP, up to 30MB each. Clean, well-lit references with the subject clearly visible work considerably better than crowded ones — the model has to work out what in the picture you meant.
Yes, one of each under Advanced, up to 30 seconds apiece, addressed as @Video1 and @Audio1 in the prompt. The video lends its motion and cuts the price by about 40%; the audio drives speech, singing or rhythm. They stack with your reference images rather than replacing them.
You are not charged. Credits are held when you submit and released back to your balance if the model fails or the provider never returns a result. Generated videos stay available for seven days, so download anything you want to keep.
Try Reference to Video
Bring the character, the location and the props. Say which is which, and let it shoot.