Baskit · video character swap
AI video character swap: replace a person in your clip, keep the performance
Baskit takes a 1–15 second clip and up to ten reference images, then re-renders the clip with the chosen person replaced — face, hair, body and outfit — while the original motion, camera moves, framing, cuts and audio stay in place. It runs on Wan 3.0 reference-based video generation.
How Baskit works
1. Upload the source clip (MP4 or MOV, 1–15 s, up to 100 MB). 2. Add reference images and tell Baskit who each one replaces — the person on the left, the driver, the one nearest the camera. 3. Optionally describe other changes (background, props, wardrobe). 4. Confirm you hold the rights to the people and brands shown and start; the result lands in your media library in a few minutes.
What stays and what changes
| Kept from the source | Replaced or edited |
|---|---|
| Body motion, gestures, timing | The named person: face, hair, body, outfit |
| Camera moves, framing, cuts | Props or background you ask to change |
| Dialogue timing and lip movement | Nothing else — unchanged areas are preserved |
Pricing and limits
Baskit is billed as Wan 3.0 output seconds: a 5-second clip at 720p is 150 BP (about $2.25); 480p and 1080p rates are shown before you start. Input clip plus output must stay within 30 seconds, which is why source clips are capped at 15 seconds. Failed jobs return their BP.
Frequently asked questions
Can I swap more than one person?
Yes. Add one reference image per person and assign each a target (left, centre, right, driver, passenger and so on). Each is replaced in every shot of the clip.
Does the original voice survive?
The prompt instructs the model to keep dialogue, timing and lip movement from the source, and Wan 3.0 renders with audio enabled. Exact soundtrack preservation is not guaranteed by the provider, so keep the original audio track to re-sync if needed.
Is this allowed?
You must have the rights to the people, brands and footage you upload. Baskit is for your own actors, licensed footage and consented likenesses, as set out in the Content Terms.