← Blog

Seedance Character Swap: Replace a Character, Keep Every Shot

A Seedance character swap puts your character in place of the lead in an existing video. The shots, cuts, camera moves and timing stay where the original has them. Here is how to do it in Riffkit's Swap mode, engine by engine. A disclosure first: we make Riffkit.

What a Seedance character swap is

ByteDance's Seedance 2.5 tutorial on BytePlus lists "replacing the video subject" under video editing, for reference videos of 4 to 30 seconds. Its Seedance 2.0 prompt guide caps video input at 15 seconds in total.

Riffkit's Swap runs those engines for you. It renders the whole video again on the source's own clock, carrying the source clip into every render: up to 15 seconds per render on Seedance 2.0 and 2.0 Fast, up to 30 seconds on 2.5. A longer source is split at its shot cuts, up to 45 seconds per job. Because the video is rendered again, the setting and wardrobe can change too (face swap vs character swap vs a full reshoot).

Step 1: set up your character

  1. In the web app, open Characters and tap + New character.
  2. Add a name, a gender, an age range and one photo (JPG, PNG, WEBP, BMP, TIFF or GIF, up to 30 MB).
  3. Use one person, front-facing, with the face large. The layout we recommend is one image with two views of the same person on a plain background. On the left, a close-up of the face: front-facing, head and shoulders, the face filling most of the panel. On the right, the full body: standing, head to toe, in the outfit you want as the character's usual look. A natural photo gives more realistic videos than an over-retouched one. A sheet of many small poses usually fails in close-ups: the output keeps the original face and at most picks up the hair.
  4. Save. The photo goes to the video platform's review, usually for under a minute, and the status turns Ready. If it reads Review failed, upload a different photo.

Use your own photo or one you have permission to use.

Step 2: replace the character in your video

  1. Open Swap.
  2. Add the clip: an upload up to 100 MB, a TikTok link or a template. Swap keeps the full length, up to 45 seconds, so a longer source is turned away, not trimmed. For a video you made in Riffkit, download it and upload it.
  3. Pick your character. Each character makes its own video, billed on its own.
  4. Optionally, write in "What to change" (more below).
  5. Pick Seedance 2.0 Fast, 2.0 or 2.5 and a resolution. The per-second rate shows before anything renders.
  6. Tap Riff it →. Each part of the source first passes the video platform's content review, which can take several minutes on a new clip. A refused clip costs nothing.

Seedance 2.0, 2.0 Fast or 2.5?

Engine Swap rate, credits per second Longest single render Your character's voice
Seedance 2.0 Fast 50 at 480p, 100 at 720p 15 s The original voice stays
Seedance 2.0 60 at 480p, 120 at 720p, 300 at 1080p 15 s The original voice stays
Seedance 2.5 90 at 480p, 180 at 720p 30 s A new voice by default on lines spoken on camera

We recommend Seedance 2.5 for swaps; a paid Swap defaults to Seedance 2.0 at 720p, and the engine picker marks 2.5 as Recommended. On 2.5, narration (heard, not seen) keeps the original voice unless "What to change" asks, for example "the narration also in the new person's voice". "Keep the original audio" keeps every voice. The new voice is AI-generated. A character's saved voice sample is used in Adapt riffs and creations, except an Adapt with Background music set to Similar BGM.

Renders bill whole seconds, rounded up (4 seconds minimum), so a split clip can bill up to 1 second more per extra render.

Write "What to change" alongside the character

The character takes the lead. Everything else goes in "What to change": a prop, an outfit, the setting, the rest of the cast. An outfit, a place or a prop can also come from a photo you pick under "Reference images". Leave both empty and only the character changes. To keep the person and change only the clothes, see how to change the outfit in a video with AI, or start from a photo of the garment; to change only their color, see how to change the color of clothes in a video. To keep the person and change only the place, see how to change the background of a video; only the hair, how to change hair color or hairstyle in a video. Not sure a re-render is the right kind of tool? The kinds of AI that replace a character in a video sets them side by side.

Here is what a saved character looks like in a swap:

The car scene side by side: the original woman on the left, our saved character in her place on the right, in the same gown and pose

Click for the silent side-by-side: our own 12-second clip on the left, the same clip with a saved character in her place on the right. Seedance 2.5 at 720p, about 5 minutes to render.

We picked the character and wrote one line that only named her. Her face is in every shot, and the shots, cuts and caption stayed. In the car and on the red carpet she wears the original's gown and gloves. One scene drifted: on the fire escape she wears the jacket from her character image and holds the phone in her hand, where the original wedges it between shoulder and ear. Run again at 480p with the box empty, her face was in every shot too.

Another example changed its cast by description, with no saved character. We reshot the 30-second STORM II clip on Seedance 2.5 at 720p as a grandma choir: grandmothers in lilac cardigans fill the crowd, and the lead becomes a grandfather holding tea where the original holds a cigarette. A 480p draft came first, at half the 720p rate; the final took about ten minutes. Lessons that carry over:

  • Name the hard moments. Young faces crept back at the climax until the brief said "for the whole video, including the final frenzy".
  • Tie props to actions. "Sipping it where the original takes a drag" put the tea on the cigarette's beats.
  • Expect misses. The crowd never mouthed the song, in two tries.

STORM II is by GENER8ION (Surkin and Romain Gavras), starring Yung Lean, with choreography by Damien Jalet. The footage and music are theirs. Ours is a non-commercial demo, and YouTube flagged it as claimed content. The full brief is in our STORM II how-to.

Someone changed only by words has no reference image, so the face can differ between shots. A saved character gives the model a photo to work from in every shot, which is also how you keep one character steady across a series.

What stays, what changes, and the limits

  • Stays: shots, cuts, camera moves, framing, poses, timing, full length, aspect ratio and language, and the source's sound outside the parts where your character speaks on camera or your text changes a voice. In those parts, on every engine, the sound is newly rendered: the voice (a new one by default on 2.5, the original one on 2.0 and 2.0 Fast) over the source's music track when Riffkit has one; the source's effects and other voices drop out there.
  • Changes: the lead becomes your character, plus whatever "What to change" names, the outfit, place or prop in any reference image you picked, and a product with images if you picked one (our product swap test). With no character picked, the original lead stays, real face included; in our words-only test, describing a new woman changed her hair, not her face.
  • Something has to change. A character counts only when the source shows someone to replace. A swap with nothing to change is refused before anything is billed.
  • Billing. Failed renders cost nothing. If a run fails partway or you cancel, the seconds already rendered are billed.
  • Sound. A render whose new sound is rejected as a copyrighted recording keeps the source clip's sound, so on 2.5 your character may speak in the original voice there.
  • Rights. Swap keeps the source's sound wherever no voice changes, music included, and someone else's TikTok still belongs to them. For another creator's winner, use Adapt, which keeps the formula and makes new footage (where the line sits). When Riffkit has the source's music as its own track, Adapt keeps that BGM by default: the music is part of why the video won, and the new video follows its beats. The track's rights stay with its owner, so check you're cleared to use it where you post.

Free accounts

Eligible new accounts start with free credits, no card needed. A free account can Swap on Seedance 2.0 Fast at 480p (the default) or Seedance 2.0 at 480p. A short clip comes back whole. Most longer ones start with a free first section, and a plan finishes the rest. Seedance 2.5, MiniMax H3, 720p and 1080p come with a plan, from $15 a month for 3,000 credits.

Try it

Set up your character, then open Swap and run it on a clip of your own. In a chat instead? Replace the person in a video from Claude or ChatGPT.

FAQ

How do I do a Seedance character swap?

In Riffkit, set up a character first. On the Characters page, add a name, a gender, an age range and one clear, front-facing photo, then wait for the video platform's review to mark it Ready, which usually takes under a minute. Then open Swap and add a clip of up to 45 seconds (an upload up to 100 MB, a TikTok link or a template). Pick the character, choose Seedance 2.0 Fast, 2.0 or 2.5, and submit. The shots, cuts, camera moves and timing stay where the original has them, and the lead becomes your character.

Can Seedance 2.0 replace a character in a video?

Yes. ByteDance's Seedance 2.0 prompt guide on BytePlus lists local replacement among its video-editing jobs and allows up to 15 seconds of video input in total. Riffkit's Swap runs Seedance 2.0 and 2.0 Fast in renders of up to 15 seconds each, split at the source's shot cuts, and one job takes a clip of up to 45 seconds at full length. On both engines the original voice stays. On every engine, a part where your character speaks on camera, or where your text changes a voice, gets newly rendered sound (the voice over the source's music track when Riffkit has one, with the source's effects and other voices dropped there), and everywhere else the sound stays the source's. Swap on Seedance 2.0 costs 60 credits a second at 480p, 120 at 720p and 300 at 1080p.

Which Seedance model is best for a character swap?

For swaps, Riffkit recommends Seedance 2.5; a paid Swap defaults to Seedance 2.0 at 720p, and the engine picker marks 2.5 as Recommended. Seedance 2.5 renders up to 30 seconds in one call. Its Swap rate is 90 credits a second at 480p and 180 at 720p. Seedance 2.0 Fast (50 credits a second at 480p, 100 at 720p) and Seedance 2.0 (60, 120, and 300 at 1080p) render up to 15 seconds per call, and free accounts can use their 480p tiers.

Does a Seedance character swap keep the original voice?

It depends on the engine. In Riffkit's Swap, Seedance 2.0 and 2.0 Fast keep the original voice. On Seedance 2.5, the new character says their on-camera lines in a new AI voice by default, and narration keeps the original voice unless you ask for a change in "What to change", for example "the narration also in the new person's voice". On every engine, a part where your character speaks on camera, or where your text changes a voice, gets newly rendered sound: the voice over the source's music track when Riffkit has one, with the source's effects and other voices dropped there. Everywhere else the sound stays the source's. Writing "keep the original audio" keeps every voice.

What photo works best for an AI character swap?

Use one clear, front-facing photo of a single face, large in the frame, on a plain background. A natural photo gives more realistic videos than an over-retouched one. The layout we recommend is one image with two views of the same person: a close-up of the face on the left and the full body on the right. A sheet of many small poses usually fails in close-ups: the output keeps the original face and at most picks up the hair. In Riffkit the photo can be a JPG, PNG, WEBP, BMP, TIFF or GIF of up to 30 MB. It passes the video platform's review before the character can be used in a swap.

Can I do a Seedance character swap for free?

If your account is eligible, you can try it in Riffkit with the free credits new accounts start with, no card needed. A free account can Swap on Seedance 2.0 Fast at 480p or Seedance 2.0 at 480p. A short clip comes back whole. Most longer ones start with a free first section, and a plan finishes the rest. Seedance 2.5, MiniMax H3, 720p and 1080p come with a plan, from $15 a month for 3,000 credits.

Keep reading

How to Add Snow or Rain to a Video with AI

Add snow or rain to a video you already have with AI: two Seedance 2.0 tests on one clip, the exact prompts, what held and what didn't.

Add an Object to a Video with AI (Real Test)

Add an object to a video with AI and keep the person, moves and sound. Our test put sunglasses and a plant into an 11-second clip, checked every half second.

Make a Christmas AI Video from a Clip You Already Have

Make a Christmas AI video from a clip you already have: five Seedance 2.0 tests with lights, a tree, pajamas, snow and a dog, the prompts, and what slipped.