HappyHorse Video Edit on fal: What It Costs to Replace a Person in a Clip, and the Limits (October 2026)
HappyHorse's video-edit endpoint is a different model from the 1.1 generators: version 1.0, billed on input plus output seconds, so $0.28 per delivered second at 720p. The limits, the audio setting, the all-faces behaviour, and how it compares with Recast and Kling O3.
When you paste a Reel into AutoHustle's autopilot and ask for your face, the model that does the replacing is HappyHorse video-edit. It has been in production since May, it has handled hundreds of runs, and it is easy to misunderstand, because it shares a brand with the HappyHorse 1.1 generators and is actually a different model. This guide documents what it is, what it really costs, and where its limits are, partly so that the comparisons with the newer candidates, Recast and Kling O3 edit, are made against facts.
It is version 1.0, not 1.1
HappyHorse 1.1 has three endpoints (text-to-video, image-to-video and reference-to-video) and none of them reads a source video. The video-edit endpoint is a separate model that is still at version 1.0. If you have read that "HappyHorse 1.1 cannot edit video", that is true of the 1.1 generators and not of this endpoint. Both facts are in the HappyHorse explainer.
What it takes
| Input | Limit |
|---|---|
| Source video | 3 to 60 seconds |
| Output | up to 15 seconds per call |
| Reference images | up to five, cited in the prompt as @Image1 to @Image5 |
| Resolution | 720p or 1080p (the endpoint defaults to 1080p) |
| Audio | an audio_setting option; origin keeps the source soundtrack |
The asymmetry to remember: you may send up to a minute of video, but you get back at most 15 seconds. A longer Reel is therefore processed in chunks, which is why our pipeline cuts a source into pieces of at most 15 seconds and joins the results.
What it really costs
The listed rate is $0.14 per unit, and the unit is a second of input plus a second of output. A 15-second clip in, 15 seconds out, is 30 billed seconds. That works out at $0.28 per delivered second at 720p and $0.56 at 1080p.
| Delivered length | 720p | 1080p |
|---|---|---|
| 5 seconds | $1.40 | $2.80 |
| 10 seconds | $2.80 | $5.60 |
| 15 seconds | $4.20 | $8.40 |
The mistake to avoid is reading the $0.14 as a per-second price: that halves the real figure, and fal's pricing data does not say what the unit measures. In the Studio the cost is quoted in credits before you press Generate, at about four credits per second of finished video.
What it does well
- Real faces. It accepts them; this is the reason it is our primary replacement model, since Seedance rejects them.
- Motion and framing. It keeps the source's camera and action, because it edits the video rather than regenerating the scene.
- The soundtrack. With
audio_settingset tooriginthe original audio survives, which matters because a trending Reel usually is its audio.
Where it struggles
All faces become one. It replaces every face it finds with the avatar you supply. A Reel with two people is a case where the result is not what you expect, which is why the Studio warns you before it runs. The case is covered in who gets replaced when there are several people.
Chunk boundaries. Every 15-second cut is a place where clothes, hair and skin can shift, and the effect compounds on a long Reel.
Identity is not guaranteed. No video model is certified for it, and our own similarity checks on difficult source clips have sometimes fallen short. That is exactly why we are testing alternatives rather than defending this one.
The candidates it now faces
| HappyHorse video-edit | Kling O3 edit | MiniMax H3 Max Recast | |
|---|---|---|---|
| Source length | 3–60 s in, 15 s out | 3–15 s | 5–30 s, shots ≤15 s |
| Price per delivered second | $0.28 (720p) | $0.14 | $0.30–0.45 (reported) |
| Who becomes whom | all faces to one avatar | via @Element1 in the prompt | one photo per person, left to right |
| Keeps the original cuts | within each chunk | within the clip | yes, natively |
| Status with us | in production | being tested | being tested |
We will switch only if a blind comparison on our own clips says a candidate is better, and we will publish what we find. Replacing a person in a video explains the choice in plain terms.
Common questions
Why is it $0.28 a second when the price says $0.14?
The unit is a second of input plus a second of output, so a clip of N seconds in and N seconds out bills 2N units. At 720p that is $0.28 per delivered second.
Is it the same model as HappyHorse 1.1?
No. The edit endpoint is version 1.0 and the 1.1 generators are separate models with their own limits.
What happens with two people?
All faces found are replaced with the one avatar you supply, so the Studio warns before running when it detects more than one.
Does it keep the soundtrack?
With audio_setting set to origin, yes. That matters because a trending Reel usually is its audio.
How to do this without code: HappyHorse video-edit is the model the autopilot uses to put you into a pasted Reel. 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: AI Video Generation Pricing Compared (October 2026): What a Second Really Costs · Kling 3.0 vs HappyHorse 1.1 for Character Replacement (August 2026)
For developers: the fal endpoint
| Model | fal endpoint |
|---|---|
| HappyHorse video-edit (1.0) | alibaba/happy-horse/video-edit |
Inputs: source video (3–60 s), up to five reference images (@Image1…@Image5), prompt, resolution, audio_setting (origin keeps source audio). Billing: $0.14 per unit, one unit per second of input plus one per second of output. Read on 18 October 2026.