Replace Character
Replacing a character in a video means putting a different person into an existing shot while the original performance stays: the movement, the timing and the camera are kept, and the body, clothing and face are regenerated as someone else. You upload the clip and a reference image of who should appear instead.
This is the heavier of our two swap tools and not always the one you want. If only the face needs to change, video face swap is faster, cheaper and holds the rest of the frame exactly. Character replacement runs today on Wan 3.0: pick Wan 3.0 in the editor, attach your clip and a photo of the new person, and describe the swap.
How to replace a character with Wan 3.0
Open the editor on this page, choose Wan 3.0, attach the clip in the Video slot and a photo of the new person in the Image slot. With a clip attached, the photo is read as the reference for who should appear, not as the opening frame. Then write the prompt. It works best in three parts: who is replaced, who replaces them, and everything that must not change.
For example, to put someone else behind the wheel of a driving shot: "Replace the driver in video 1 with the person in figure 1. Keep their face, hair, build and clothing as in figure 1. Keep everything else from video 1 unchanged: the car, the interior, the road, the lighting, the camera angle and movement, the driving motion and its timing." Name the things you care about — anything you leave out, the model may change.
Pick an output length close to your clip’s length. Auto length is switched off while a clip is attached, because the clip’s seconds and the output’s seconds together may not exceed 30. If the source has captions or a watermark you do not want carried over, add a last sentence asking for on-screen text to be removed.
| Resolution | Standard | Prime (about 6x faster) | 8 s clip + 8 s output, standard |
|---|---|---|---|
| 480P | 7 credits / s | 11 credits / s | 112 credits |
| 720P | 14 credits / s | 22 credits / s | 224 credits |
| 1080P | 29 credits / s | 45 credits / s | 464 credits |
Replace character or face swap?
These two jobs get confused constantly, and picking the wrong one wastes both time and credits. The difference is how much of the frame gets rebuilt.
Face swap replaces the face region and composites it back onto the original footage. Everything else — hair, body, clothing, hands, background — is your untouched original. It is the right answer whenever the body in the shot can stay as it is.
Character replacement regenerates the whole person. That is necessary when the new character has a different build, different clothing, or is not a person at all, and it is overkill when they do not.
| Face swap | Replace character | |
|---|---|---|
| Changes | The face only | Face, body, hair and clothing |
| Keeps | Body, clothing, background untouched | Movement, timing and camera |
| Use when | The body in the shot works as-is | Build or wardrobe must change too |
| Cost | From 5 credits per segment | Per second of clip + output (Wan 3.0) |
| Status | Running today | Running today on Wan 3.0 |
What the reference image needs
The reference decides who appears, and a weak reference is the most common cause of a disappointing swap. The model needs to understand the person from one picture, so give it the clearest possible statement of who they are.
- A full-body shot when the clip shows a full body. A head-and-shoulders reference forces the model to invent the build, and it will invent a generic one.
- Even, front-facing light. Heavy shadow on one side means the model guesses at half the face.
- Neutral expression and an unobstructed face — no sunglasses, no hand across the jaw, no hair over the eyes.
- Similar framing to the target shot where you can manage it. A reference standing upright matches a clip of someone standing far better than a seated portrait does.
What survives the swap and what does not
The performance survives: where the subject goes, when they turn, how fast they move, how the camera follows. That is the whole reason to use this rather than generating a new clip from scratch — you are keeping choreography that would be hard to specify in a prompt.
Fine identity detail does not survive reliably. The new character is recognisably the person in your reference in a good result and recognisably a relative of them in a mediocre one, and the difference between those two outcomes is mostly reference quality and clip length.
Hands and contact points are the weakest area, as they are in every generative video tool. A shot where the subject handles an object is much harder than one where they walk across frame.
When not to use this
- Putting a real person into footage they were never in. Use this with the consent of the person in the reference image — this is the limit that matters most on this page, and it is not negotiable.
- Anything intended to deceive: fake testimony, fabricated evidence, a public figure appearing to do something they did not.
- Shots where only the face needs changing. Video face swap keeps the rest of your footage intact, costs less and works today.
- Clips where the subject handles objects closely. Hands at contact points are the least reliable part of the output.
- Long takes. As with everything generative here, the last seconds are the least stable.
- Crowd scenes. The model targets one subject; several people in frame produces unpredictable results.
Questions people ask
What is character replacement in video?
It regenerates the person in an existing shot as somebody else while keeping the original movement, timing and camera. You supply the clip and one reference image of the replacement, and the model rebuilds the body, clothing and face to match the reference while following the original performance.
How is this different from face swap?
Face swap changes only the face region and composites it onto your untouched footage, so the body, clothing and background are exactly as you filmed them. Character replacement regenerates the whole person. Choose face swap unless the build or the wardrobe also has to change — it is cheaper, more faithful to the rest of the frame, and it works today.
What makes a good reference image?
A sharp, evenly lit, front-facing shot with the face unobstructed, framed the way the clip frames its subject — full body for a full-body clip. A head-and-shoulders reference used on a full-body shot leaves the model to invent the build, and it invents a generic one.
Will the replacement look exactly like my reference?
Close, not exact. Expect a recognisable likeness rather than an identical one, and expect it to hold better over a short clip than a long one. If exact likeness is essential, this is not the right tool.
Can I replace someone in a video I did not film?
Only if you have the rights to the footage and the agreement of the person whose likeness you are adding. Both conditions, not one.
How much does it cost?
On Wan 3.0 you pay per second for the clip you upload plus the video you get back, at the same rate: 7 credits a second at 480P, 14 at 720P and 29 at 1080P on standard, more on the faster prime mode. An 8-second clip turned into an 8-second result at 720P is 16 seconds, or 224 credits. The exact figure is on the Create button before you spend anything.
Related tools
Checked 30 September 2026.