Bring a clip you already have and change what is inside it. Replace a person with another character, swap a product, change the setting, or keep the shot running past where it stopped, with no queue and no watermark on the download. New accounts start with free credits.
Real output
One clip went in as the source and one still went in as the replacement reference. A single sentence of instruction produced the result: same camera, same lighting, new object.
Prompt
Change the perfume in the video gift box to the face cream from the image.
Source clip, plus the reference still that replaces the product.
Input video
Reference image
Result
Seedance output, watermark-free.
Use cases
Video to video is the mode for footage that is almost right: the shot works, but something inside it has to change.
01
Keep the set, the lighting and the camera move, and swap who or what is in the frame: another character from a still, the SKU that actually shipped, a different car. One clip can then cover a whole catalogue or a whole cast.
02
Push the same take into a different look, season, palette, or era while the blocking and the timing stay exactly where they were.
03
Continue a clip past its last frame when the edit needs one more beat and the original footage is long gone.
04
Change one region of the frame, such as a label, a sign, or a garment, and leave everything around it untouched.
How it works
Attach the clip and the composer switches itself. Adding a still alongside it turns the pair into a reference group.
01
Drop the clip in first. Stills added next join it as reference material rather than replacing it, and on the Seedance models more clips can follow: up to three on Seedance 2.0 and ten on Seedance 2.5.
02
Say what should be different. Replace the perfume in the gift box with the cream from the image is the shape that works; a full scene description fights the source.
03
The result lands next to the original in your asset library, watermark-free, so you can judge the swap frame by frame.
Jobs people arrive with
These are the jobs the clip arrives with, named the way people ask for them. Each has a direct answer, the steps in this composer, the instruction verbatim and the model that fits, so you can paste the prompt and run it on your own footage.
01
Yes. Attach the clip, add one still of the character who should appear instead, and name the swap in a sentence. The model keeps the movement, timing, camera and lighting of the original and re-renders the person. On free credits this runs on Seedance 2.0 Mini at up to 720p; Seedance 2.5, on a paid plan or a credit pack, does it as a localized edit that leaves the rest of the frame untouched. One limit to know before you start: the Seedance models refuse some footage of real people and photorealistic faces under ByteDance’s content rules, so a stylized or clearly fictional replacement is the reliable choice, and a refused job is not charged.
How to do it
Prompt
Replace the man in the video with the character from the image. Keep his movements, timing, framing, camera motion and lighting exactly as they are; change nothing else in the scene.
Settings
Prompt recipe: no clip attached, paste it and run it.
02
Yes, and it is the job this entry does best, because what you keep is the shot and what you change is small. Give the model one still of the replacement, name the object to remove and the image it comes from, and leave the rest of the sentence empty. The perfume-to-face-cream pair in the demo above was made exactly this way on Seedance 2.0: one clip, one still, one sentence.
How to do it
Prompt
Replace the glass on the table with the can from the image. Keep the hand, the table, the lighting and the camera movement unchanged.
Settings
03
Yes, without a green screen. Describe the new place and say that the people, their movement and the camera stay. The model re-renders the environment around the action: a kitchen becomes a studio, a car park becomes a coastline, noon becomes night. On Seedance 2.5 this is a localized edit that keeps the subject stable; on the Seedance 2.0 family, including Mini on free credits, the scene is re-rendered from the clip, which suits a full restyle more than a one-object fix.
How to do it
Prompt
Keep the two people, their movements and the camera path exactly as in the source. Change the setting to a rooftop at dusk with city lights behind them; match the new light on their faces and clothes.
Settings
Prompt recipe: no clip attached, paste it and run it.
04
Yes. The instruction names what appears, where, and how it behaves, and the rest of the shot is left alone: twenty students in an empty classroom, a car in an empty driveway, a product on a bare table. A reference still of the thing to add makes it specific; without one the model invents it from the words. Crowds and many small additions are harder than one clear object, so start with one.
How to do it
Prompt
Add a red sports car parked in the empty driveway on the left, matching the afternoon light and shadows. Keep the house, the camera movement and everything else exactly as in the source.
Settings
Prompt recipe: no clip attached, paste it and run it.
05
Yes. This is the smallest kind of replacement: the person, the performance and the camera stay, and one attribute changes. Name the attribute and the new value, and say the face and the movement stay. A reference still of the garment or hairstyle helps the model match a specific look. The same real-people caveat applies: footage with photorealistic faces can be refused, and a refusal is not charged.
How to do it
Prompt
Keep my face, expressions, movements and the room exactly as in the source. Change only my hairstyle to soft natural curls, shoulder length.
Settings
Prompt recipe: no clip attached, paste it and run it.
What you can generate
Only the models that accept a reference clip appear in this table, and every value is read from the registry the composer enforces. The inputs column separates the two ways a clip can be used: as a reference that the scene is re-rendered from, and, on Seedance 2.5 alone, as the source of a dedicated edit or extend job.
| Model | Max resolution | Clip length | Audio | Inputs | On free credits | Starting cost |
|---|---|---|---|---|---|---|
| Seedance 2.5 | 1080p | 4–30s | On/off toggle | Up to 10 reference clips · Edit and extend modes · Up to 30 reference images | Paid plan or pack | From 12 credits (4s · 480p) |
| Seedance 2.0 | 1080p | 4–15s | On/off toggle | Up to 3 reference clips · Up to 9 reference images | Paid plan or pack | From 9 credits (4s · 480p) |
| Seedance 2.0 Fast | 1080p | 4–15s | On/off toggle | Up to 3 reference clips · Up to 9 reference images | Paid plan or pack | From 7 credits (4s · 480p) |
| Seedance 2.0 Mini | 720p | 4–15s | On/off toggle | Up to 3 reference clips · Up to 9 reference images | Yes | From 5 credits (4s · 480p) |
| Kling O3 | 1080p | 4–15s | On/off toggle | One reference clip | Paid plan or pack | From 10 credits (4s · 720p) |
| Kling 3.0 Motion Control | 1080p | Follows the reference clip (3–30s) | No audio option | One reference clip | Paid plan or pack | From 9 credits (3s · 720p) |
| Kling 2.6 Motion Control | 1080p | Follows the reference clip (3–30s) | No audio option | One reference clip | Paid plan or pack | From 6 credits (3s · 720p) |
| MiniMax H3 | 2K | 5–15s | No audio option | Up to 3 reference clips · Up to 9 reference images | Paid plan or pack | From 23 credits (5s · 2K) |
| Wan 3.0 | 1080p | 2–30s | On/off toggle | Up to 5 reference clips · Up to 10 reference images | Paid plan or pack | From 2 credits (2s · 480p) |
The plan sets the ceiling and the model sets what is possible under it. A 4K-capable model still renders at 720p on free credits, and Seedance tops out at 1080p on every plan.
| Plan | Seedance models | Other models |
|---|---|---|
| Free credits | 720p | 720p |
| Lite | 720p | 1080p |
| Pro | 1080p | 2K |
| Premium | 1080p | 4K |
Last updated
Worked examples
The first example is the real pair from the demo above with its unedited instruction. The other two are prompt recipes for the two remaining jobs this entry does, a restyle and an extension, written against Seedance 2.5’s modes so you can run them on a clip of your own.
01
Before
A finished five-second gift-box clip with a perfume bottle inside it, plus one still of the face cream that should be there instead. Two attachments, one sentence.
After
The same camera move, set and lighting with the cream in the box. Nothing else in the frame was mentioned, so nothing else moved. This is the pair shown in the demo section above, and the shape of the instruction is the lesson: name the object to replace and the image it comes from.
Prompt
Change the perfume in the video gift box to the face cream from the image.
Settings
02
Before
A handheld eight-second walkthrough of a furnished living room shot at noon: flat light, correct blocking, the kind of take you would keep if only the mood were different.
After
The same walk with the same timing, now at dusk: table lamps on, cool blue window light, longer shadows, with the furniture, layout and materials unchanged because the instruction told the model to leave them alone.
Prompt
Keep the camera path, framing and timing of the source clip exactly. Change the time of day to dusk: warm table lamps switched on, cool blue light through the windows, soft long shadows. Furniture, layout and materials stay the same.
Settings
Prompt recipe: no clip attached, paste it and run it.
03
Before
A six-second clip of a cyclist crossing a bridge at golden hour that cuts just as she reaches the far side, one beat before the edit needs it to.
After
The shot continues: she rides off the bridge into a narrow town street while the camera holds its pan at the same speed, and the light and grain match the original so the join does not read as a cut.
Prompt
Extend the video by 5 seconds. The cyclist rides off the far end of the bridge into a narrow town street, the camera keeps panning at the same speed, same golden-hour light and film grain, no cuts.
Settings
Prompt recipe: no clip attached, paste it and run it.
Questions
Choosing an entry
The three scene pages share one composer. What you attach decides the entry, and the model you pick decides the ceiling.
A clip is the right entry when the footage is almost right and only one thing in it has to change. When the shot does not exist at all, start from text to video and describe it. When you have a packshot, a poster or a portrait but no footage, start from image to video, where the still becomes the first frame and the prompt supplies the motion.
Within this entry the model matters more than anywhere else. Seedance 2.5 is the only model with dedicated edit and extend modes, and the only one that takes ten reference clips and 30 reference images, so localized fixes and extensions belong to it. Seedance 2.0 and its Fast and Mini tiers re-render a scene from a reference clip at a lower cost, which is enough for a restyle or a product swap. When a performance from one clip has to land on a still of someone else, use motion control: Kling 2.6 Motion Control does it at the lower credit rate, and Kling 3.0 Motion Control is the newer generation.
When the clip is right but ends too early, you do not need to change anything inside it: the AI video extender continues it from the last frame for 4 to 30 seconds on Seedance 2.5, with a worked example, the exact instruction and the cost per plan.
Keep going
Take the clip further with another model, or check what your credits cover.
Free credits
Attach a clip, describe the one thing that should be different, and download the result without a watermark. Free credits on a new account.
Edit a video