Happy Horse Video Edit
Happy Horse Video Edit (HappyHorse-1.0 video edit, model id happyhorse-1.0-video-edit) is the editing model in Alibaba's Happy Horse 1.0 video family, built by the Alibaba Token Hub (ATH) unit. Happy Horse 1.0 first appeared anonymously on the Artificial Analysis video arena in April 2026, and Alibaba confirmed it was theirs on 9 April; the video edit endpoint went live on fal, as an official API partner, on 27 April, and is also on Alibaba Cloud Model Studio. It takes an existing clip of 3 to 60 seconds plus a text instruction and up to five reference images, and rewrites the footage: global changes such as style, season or lighting, or local ones such as an object's colour or a person's clothes, while keeping the source's framing and motion. The output is as long as the source, up to 15 seconds, at 720p or 1080p, and it can keep the original sound track. Happy Horse 1.1 followed in June for generation; the edit model is still 1.0.
Anvisha Pai, Co-founder & CEO, Voyager
Alibaba · released
›What we asked for
Turn this into a snowy winter morning: fresh snow on the roofs and the boats, overcast soft light. Keep the camera move and the town exactly as they are.
alibaba/happy-horse/video-edit
resolution: 720p, audio_setting: origin
returned 1280 by 720 pixels
At a glance
- Price
- On fal at 720p, $0.14 per second of source plus $0.14 per second of output, so $0.28 a second for a same-length edit; double at 1080p. Our five-second 720p edits cost about $1.41 each.
- Length
- The output is as long as the source, from 3 up to 15 seconds. Sources can run to 60 seconds, but only the first 15 are edited.
- Sound
- audio_setting=origin keeps the source track; ours came back identical to the input on all three clips. The default, auto, lets the model decide whether to regenerate it.
- References
- Up to five images, cited in the prompt as @Image1 to @Image5, for swapping in a face, an object or a style. Not used in our runs.
- Resolution
- 720p or 1080p, keeping the source's aspect ratio; fal defaults to 1080p. Our runs were at 720p.
- Measured wait
- 145 to 194 seconds per five-second 720p edit, three submitted together on fal. One batch, not a service guarantee.
Anvisha's take
The season change is the reason to use it. Our golden-hour harbour drone shot became an overcast winter morning with snow on every roof, quay and boat, and the camera move, the fort and the layout of the town held frame for frame. The recolour worked too, the dripper is a deep green glaze on the same turntable, but the glaze carries painted-looking reflections and the spout curls a little more than it did. The wardrobe edit shows the limit: the jumper is yellow, but the white shirt collar went with the old one, and her face drifted to a rounder, slightly different woman despite 'keep her face' in the prompt. With audio_setting set to origin the sound came back identical to the source every time. At about $1.41 for a five-second 720p edit and two and a half to three minutes of waiting, I would use it for scene-wide restyles and object colour, and not for anything where a person's face has to survive.
Happy Horse Video Edit: real results on our 3 standard video edit tests
Happy Horse Video Edit works on something you give it, so every model that does this job gets the same 3 video edit tests: the same clip and the same instruction. Each result sits next to what went in. These are the clips Happy Horse Video Edit returned, one try each, nothing re-rolled or retouched. Each tile shows one frame of the clip until you press play. Generated on . The 3 timed clips took 145 to 194 seconds each and cost $4.24 in total.
“Change the coffee dripper to a deep forest green glazed ceramic. Keep the turntable, background, lighting and motion exactly as they are.”
- 720p
›What we asked for
Change the coffee dripper to a deep forest green glazed ceramic. Keep the turntable, background, lighting and motion exactly as they are.
alibaba/happy-horse/video-edit
resolution: 720p, audio_setting: origin
returned 1280 by 720 pixels
“Turn this into a snowy winter morning: fresh snow on the roofs and the boats, overcast soft light. Keep the camera move and the town exactly as they are.”
- 720p
›What we asked for
Turn this into a snowy winter morning: fresh snow on the roofs and the boats, overcast soft light. Keep the camera move and the town exactly as they are.
alibaba/happy-horse/video-edit
resolution: 720p, audio_setting: origin
returned 1280 by 720 pixels
“Change her top to a bright yellow knitted jumper. Keep her face, her words, her movements and the kitchen exactly as they are.”
- 720p
›What we asked for
Change her top to a bright yellow knitted jumper. Keep her face, her words, her movements and the kitchen exactly as they are.
alibaba/happy-horse/video-edit
resolution: 720p, audio_setting: origin
returned 1280 by 720 pixels
How the cost under each clip was worked out: fal's Happy Horse video edit rate at 720p, $0.14 per input second plus $0.14 per output second (fal model page, read 2026-10-28), the output taken as long as the source for the 5.04-second source clip; fal does not return the charge.
Happy Horse Video Edit pricing
| Where | Unit | Price | Source |
|---|---|---|---|
| fal | Second at 720p, source and output each fal's playground note charges $0.14 per input second and $0.14 per output second, $0.28 a second for a same-length edit. The same page's pricing section says billing is on output length only; we costed our runs on the higher reading. | $0.14 | fal model page, read |
| fal | Second at 1080p, source and output each $0.56 a second for a same-length edit on the playground's reading. 1080p is fal's default. | $0.28 | fal model page, read |
Is Happy Horse Video Edit free?
No free tier on fal: it charges per second with no minimum. Alibaba's own Model Studio API bills by the output's duration; its API reference does not give the rate, and we did not confirm a free quota for this model.
- •fal: pay per second, no subscription, no free allowance listed.
- •Alibaba Cloud Model Studio: billed on the generated video's duration; rates on Alibaba's pricing pages, which we did not verify.
fal model page, read
Where to use Happy Horse Video Edit
- fal · Hosted API and playground; the endpoint used for the edits below.
- Alibaba Cloud Model Studio · Alibaba's own async API, in Beijing, Singapore, Virginia, Frankfurt and Tokyo; adds a 'Happy Horse' watermark unless you turn it off.
- Voyager · Runs inside Voyager, the agent for creative work, on Voyager credits on every paid plan, or with your own fal key.
Voyager
Run Happy Horse Video Edit inside Voyager
Voyager is an agent for creative work that runs every model in this reference, Happy Horse Video Edit included. Brief it, and it picks the model, makes the thing and hands back something you can edit. Create an account to get started.
What we noticed running Happy Horse Video Edit
- •A deep green glaze on the same turntable, with the base holes and the rotation kept. The matte surface turned glossy, as 'glazed' asked, but the reflections look painted on as nested cone shapes, and the spout curls more than in the source. See the recolour a product clip.
- •Golden-hour haze replaced by overcast winter light, with snow on the roofs, quays and boats; the drone move, the fort and every building stay where they were, frame for frame. See the change the season clip.
- •A bright yellow crew-neck jumper, but the white shirt collar disappeared with the old jumper, and her face drifted: rounder and slightly different, though the mouth still moves with the original words. See the change clothing clip.
- •With audio_setting=origin the sound track came back identical to the source on all three clips, and each output matched its source's length to the frame.
Happy Horse Video Edit limits and API parameters
| Limit | Value |
|---|---|
| Source clip | MP4 or MOV (H.264 recommended), 3 to 60 seconds, up to 100 MB, above 8 fps, aspect between 1:2.5 and 2.5:1. On fal the longer side is at most 2160 pixels and the shorter at least 320. fal model page, read |
| Output length | 3 to 15 seconds: a source of 15 seconds or less keeps its length; a longer one is cut to its first 15 seconds. |
| Prompt | 2,500 characters on fal. Alibaba's own API takes up to 5,000 non-Chinese or 2,500 Chinese characters and truncates the rest. |
| Reference images | Up to 5, at least 300 pixels a side, aspect between 1:2.5 and 2.5:1; 10 MB each on fal, 20 MB on Alibaba's API. |
›Every setting, for developers
fal (alibaba/happy-horse/video-edit)
fal model page, read
| Parameter | Type | Values | Default |
|---|---|---|---|
| video_url | string | URL of an MP4 or MOV Required. 3 to 60 seconds, up to 100 MB. | |
| prompt | string | Up to 2,500 characters Required. Cite references as @Image1 to @Image5. | |
| reference_image_urls | array | Up to 5 URLs, JPEG, PNG or WEBP At least 300 pixels, up to 10 MB each. | |
| resolution | enum | 720p, 1080p | 1080p |
| audio_setting | enum | auto, origin origin keeps the source's sound; auto lets the model decide whether to regenerate it. | auto |
| seed | integer | 0 to 2,147,483,647 | |
| enable_safety_checker | boolean | true or false Moderates input and output. | true |
What Happy Horse Video Edit will not make
fal runs a safety checker on both the source clip and the output by default (enable_safety_checker), and lists the model for commercial use with full rights on outputs. Alibaba's own API adds a visible 'Happy Horse' watermark by default. Neither page publishes a detailed list of refused content.
fal model page, read
Alternatives to Happy Horse Video Edit
Frequently asked questions
What is Happy Horse Video Edit?
Alibaba's instruction-based video editor from the Happy Horse 1.0 family, live on fal since 27 April 2026 and also on Alibaba Cloud Model Studio. It rewrites an existing clip from a text instruction and optional reference images, keeping the framing and motion, and returns up to 15 seconds at 720p or 1080p.
How much does Happy Horse Video Edit cost?
On fal, read on 29 October 2026, $0.14 per second of source plus $0.14 per second of output at 720p, and twice that at 1080p. A five-second 720p edit comes to about $1.41. The pricing section of the same page says output seconds only, so check your bill.
How long can a Happy Horse edit be?
The output is as long as the source, up to 15 seconds. Sources can be 3 to 60 seconds; anything past 15 seconds is cut.
Does it keep the original sound?
With audio_setting set to origin, yes: on our three clips the output's sound was identical to the source's. The default, auto, lets the model decide whether to regenerate it.
Does it keep faces?
Not reliably. On our wardrobe edit the clothes changed as asked but the woman's face drifted to a slightly different person, even though the prompt asked to keep it.
How this page is made
Prices, limits and settings are read from the linked vendor and host pages on the dates shown. Every picture was generated by us through fal with the request shown under it, and the original files are kept unedited. A generation that failed is shown as a failure. Model and vendor names are used to identify the products; no vendor imagery appears on this page.
