How to Replace a Person in a Video With AI: Three Models, What Each Costs, and the Mistakes That Break It (October 2026)
Replacing the person in an existing clip while keeping the movement, the cuts and the sound is a video-edit job. HappyHorse video-edit, Kling O3 edit and MiniMax Recast compared on length limits, cost per delivered second and how they decide who becomes whom.
"Replace the person in this video with me" sounds like face swapping, and the cheapest tools do treat it that way. The version that survives full-screen viewing is different: the person is regenerated rather than composited, so body, clothes and movement all follow your references while the original scene, camera and cuts stay. That is a video-edit job, and in October 2026 three models can do it through fal. This guide compares them on what matters (length, cost, control and risk) and lists the mistakes that break the result whichever you choose.
First decide whether it is the right job
| What you want | The right tool |
|---|---|
| The same video, but with you in it, cuts and sound kept | video edit (this guide) |
| The same dance or movement on your character, new framing | motion transfer |
| A new shot with you in it, inspired by the idea | reference-to-video |
| Only the face on a talking head, short clip | a face swap tool, with its limits (face-swap tiers compared) |
If the Reel is a dance with one continuous shot, motion transfer is usually better; motion transfer or Recast draws the line.
The three video-edit models
| HappyHorse video-edit (1.0) | Kling O3 edit | MiniMax H3 Max Recast | |
|---|---|---|---|
| Source length | 3–60 s in, 15 s out per call | 3–15 s | 5–30 s, no shot over 15 s |
| References | up to 5 images | up to 4, with @Element1 syntax | 1–4 photos, one per person |
| Price per delivered second | $0.28 at 720p | $0.14 | $0.30–0.45 (reported) |
| 15-second result | $4.20 | $2.10 | $4.50–6.75 |
| Who becomes whom | every face becomes one avatar | named in the prompt | photo order, left to right, overridable |
| Cuts kept | in each chunk | within the clip | natively |
| Status with us | in production | being tested | being tested |
Prices read on 18 October 2026; Recast's are third-party figures. Details: HappyHorse video-edit, Kling O3 edit, Recast.
Step by step
- Pick a source you can use. A public Reel or a video of your own. Ownership of the footage, the soundtrack and the people in it are three separate questions (can you remake someone else's Reel).
- Check the shape of the source. One main person, mostly visible face, moderate motion is the easy case. A crowd, a mask, hands covering the face or extreme camera shake is hard for any model.
- Prepare references. A frontal photo, a three-quarter and a profile, evenly lit, whole head in frame, no sunglasses. If the outfit matters, add a full-body frame (how many photos).
- Choose the length. Shorter is safer: identity drift grows with duration and motion. A five-second test costs $0.70 to $1.50 across the three models.
- Describe what to change and what to keep. "Replace the presenter with the person in the reference, keep the kitchen, camera and audio."
- Judge at full screen with sound, on a phone. Look at the face in the last second, at the hands, and at the moment of each cut.
The mistakes that break it
- Several people and no instruction. A model that replaces "the person" may replace all of them. See who gets replaced.
- A long source in one call. Anything past the model's limit is cut into chunks, and every chunk boundary is a place for clothes and hair to shift.
- One reference photo. The model invents the profile.
- Expecting a face swap. Wardrobe and proportions come from your references and the prompt, not from the person in the source.
- Skipping rights. Platforms and rights holders act on likeness and audio, not on how the clip was made.
Honest status
None of the three has been certified for identity by anyone, and the newer two have not yet been through our own blind test. The numbers are prices and limits, not quality rankings. We run a blind three-way test on clips we own, with each run approved separately because each costs money, and will change what the Studio uses only if a candidate wins.
Common questions
Is replacing a person the same as a face swap?
No. A face swap composites a face onto the existing body. Replacement regenerates the person from references, so build, clothes and movement follow your references while the scene, camera and cuts are kept.
How long a clip can I replace?
It depends on the model: 15 seconds of output per call for HappyHorse video-edit, 3 to 15 seconds for Kling O3 edit, 5 to 30 seconds for Recast. Longer sources are processed in chunks.
Which is cheapest?
On listed prices Kling O3 edit at $0.14 per second, then HappyHorse video-edit at $0.28 and Recast at a reported $0.30 to $0.45. Cheapest only matters if the face holds, which we are testing.
Will the original sound be kept?
Yes on all three if you ask for it: audio_setting origin on HappyHorse, keep_audio on Kling, and Recast keeps the source audio by default.
How do I check the result?
Watch it at full screen on a phone with sound. Look at the face in the last second, at the hands, and at each cut.
How to do this without code: Replacing a person in a pasted Reel is exactly what the autopilot does. In AutoHustle this is the autopilot: paste the Reel link, add one photo of yourself and confirm the plan (about 4 credits per second of finished video, so 60 credits for a full 15-second call; analysing the Reel is free). See How to put your face in any Reel and Recreate an Instagram Reel.
Related reading: How to Put Your Face in Any Instagram Reel (2026 Walkthrough) · Kling 3.0 vs HappyHorse 1.1 for Character Replacement (August 2026)
For developers: the fal endpoints
| Model | fal endpoint |
|---|---|
| HappyHorse video-edit (1.0) | alibaba/happy-horse/video-edit |
| Kling O3 Pro video-to-video edit | fal-ai/kling-video/o3/pro/video-to-video/edit |
| MiniMax H3 Max Recast | minimax/h3-max/recast |
Prices as read or reported on 18 October 2026. HappyHorse bills $0.14 per unit, one unit per second of input and per second of output.