# HappyHorse: Alibaba's AI video model — what HappyHorse 1.0 and 1.1 can do

> HappyHorse is Alibaba's video model and the happyhorse.com service built on it. Version 1.0 came out in April 2026, and 1.1 on June 23, 2026. It makes 3–15 second clips at 480p–1080p with sound in one pass — from text, a first frame or up to 9 references — and the 1.0 model also edits finished videos. HappyHorse 1.0 ranks 10th on Arena. HappyHorse is available in the AI2Face generator.

Developer: Alibaba · Released: 2026-06-23 · Updated 2026-10-28
URL: https://ai2face.com/models/happyhorse

## Key facts

| | |
|---|---|
| Developer | Alibaba |
| Versions | HappyHorse 1.0 (April 2026), HappyHorse 1.1 (June 23, 2026) |
| Modes | Text, first frame, references; video editing — 1.0 model |
| Duration | 3–15 seconds, 5 by default; 24 fps |
| Resolution | 1.1 — 480p, 720p, 1080p; 1.0 — 720p, 1080p |
| Audio | Generated with the video in every mode |
| References | Up to 9 images, at least 400 px on the short side, up to 20 MB |
| Aspect ratios | 16:9, 9:16, 4:3, 3:4, 4:5, 5:4, 1:1, 21:9, 9:21 |
| Where it runs | happyhorse.com, Alibaba Cloud Model Studio API |
| In AI2Face | Yes, in the generator |

## What is HappyHorse

HappyHorse is a family of Alibaba video models. You can use them through happyhorse.com and the Alibaba Cloud Model Studio API. It's Alibaba's second video line alongside [Wan](/models/wan): Wan handles long all-in-one clips, HappyHorse handles short cinematic clips with sound.

**HappyHorse 1.0** came out in April 2026. Alibaba lists where it was put to work: short dramas, e-commerce ads, brand marketing and CG game trailers.

**HappyHorse 1.1** came out on June 23, 2026. It's an update of the same model: better motion, prompt adherence, image quality and audio sync.

On the Arena leaderboard as of September 21, 2026, HappyHorse 1.0 ranks 10th of 48 video models, right behind Meta's Muse Video.

## What HappyHorse can do

**Video with sound in one pass.** In every mode, HappyHorse generates the picture and the sound together — ambience, effects and characters' speech. Clips run 3–15 seconds at 24 frames per second.

![A film still: two friends talking at a kitchen table in the evening, a girl with curly hair gesturing with a mug, a guy in glasses listening and smiling](https://ai2face.com/_astro/dialogue.DUQrV0jH_17zVHB.jpg "A typical starting frame for a dialogue: two characters in a medium shot, lines set in the prompt. Made with Nano Banana Pro")

**Three ways to start:**

- **from text** — the whole scene comes from the prompt, and you pick the aspect ratio;
- **from a first frame** — one image becomes the opening of the clip, and the ratio follows it;
- **from references** — 1 to 9 images: characters, products, locations.

**Numbered references.** In reference mode, images are called out in the prompt in order: "the woman in the red dress from [Image 1]." In 1.1, Alibaba specifically improved how the model handles multiple references: products, characters and places stay closer to the sources.

![A 3×3 reference board: a woman in a red silk dress from three angles, a man with a grey beard in a suit, a perfume bottle, a gold watch, a marble hotel lobby, a rooftop terrace, a vintage car](https://ai2face.com/_astro/refs.Bc9Szhei_Xgmfx.jpg "Nine references — HappyHorse's maximum: two characters, two products and three locations for an ad. Made with Nano Banana Pro")

**Complex motion.** 1.1 reworked motion modeling and temporal consistency — complex action scenes come out smoother.

![A freestyle skier in an orange jacket mid-air over a snowy kicker, powder flying from the skis, bright blue sky](https://ai2face.com/_astro/action.D7WfG4nh_ZrWqW0.jpg "Fast motion — the kind of scene Alibaba uses to show off 1.1's improvements. Made with Nano Banana Pro")

**Video editing.** The `happyhorse-1.0-video-edit` model changes a finished clip from a text instruction: style transfer, element replacement. Input and output clips are 3–15 seconds at 720p or 1080p.

## HappyHorse 1.0 vs 1.1

| | HappyHorse 1.0 | HappyHorse 1.1 |
|---|---|---|
| Released | April 2026 | June 23, 2026 |
| Resolution | 720p, 1080p | 480p, 720p, 1080p |
| Duration | 3–15 s | 3–15 s |
| Modes | text, first frame, references, editing | text, first frame, references |
| Multiple references | baseline | holds products and characters better |
| Motion and audio | baseline | smoother motion, tighter sync |
| Arena rank (Sep 21, 2026) | 10 of 48 | not ranked yet |

## What HappyHorse is good for

- **Short dramas and sketches** — dialogue scenes up to 15 seconds.
- **Product ads** — the product and a character from references in one scene.
- **Brand films** — cinematic footage with sound.
- **Game trailers** — Alibaba names CG trailers as a core use case.
- **Editing finished video** — restyling or swapping an element with the 1.0 model.

## HappyHorse vs alternatives

| | HappyHorse 1.1 | [Wan 3.0](/models/wan-3-0) | [MiniMax H3](/models/minimax-h3) | [Kling 3.0](/models/kling-3-0) |
|---|---|---|---|---|
| Developer | Alibaba | Alibaba | MiniMax | Kuaishou |
| Length | 3–15 s | up to 30 s | 4–15 s | up to 15 s |
| Resolution | up to 1080p | up to 1080p | up to 2K | up to 4K |
| References | up to 9 photos | 10 photos, 5 videos, 5 audio | 9 photos, 3 videos, 3 audio | up to 7 images or elements |
| Audio | yes | yes | stereo | yes |
| Arena rank (Sep 21, 2026) | 10 (1.0) | 6 | 8 | — |
| In AI2Face | yes | yes | yes | yes |

HappyHorse is simpler than its neighbors: images only as input and up to 15 seconds. In return, prompts are shorter and the result is a cinematic clip with sound. If you need video and audio as references, go with MiniMax H3 or Wan 3.0.

## How to write prompts for HappyHorse

1. **Refer to references by number:** "[Image 1]," "[Image 2]" — in upload order.
2. **Describe the scene, the action and the camera,** and let the images supply the look — don't repeat it in words.
3. **Put lines in quotes,** and say who's speaking.
4. **Describe sound in a separate sentence:** ambience, effects, music.
5. **Keep prompts under 5,000 characters**; longer ones get truncated.

An example for the reference board above:

```
The woman from [Image 1] in the red dress steps out of the black car from [Image 9]
in front of the marble lobby from [Image 7]. The man from [Image 4] meets her and hands
her the perfume bottle from [Image 5]. She smiles: "You remembered." The camera slowly
circles them; the watch from [Image 6] glints on his wrist. Sound: soft jazz, evening city noise.
```

## How to use HappyHorse in AI2Face

HappyHorse is in the AI2Face generator next to [Wan 3.0](/models/wan-3-0), [MiniMax H3](/models/minimax-h3) and [Kling](/models/kling).

1. Sign up — new accounts get starter credits.
2. Open the generator, switch to video and pick HappyHorse.
3. Upload a frame or references, write a prompt, set the length, aspect ratio and resolution.
4. The credit cost is shown before you start. Credits don't expire.

You can make the references in the same place — characters and products on a white background for a board like the one above take one prompt in [Nano Banana Pro](/models/nano-banana-pro).

## Limitations

- **15 seconds and 1080p** are the ceiling for both versions.
- **Images only as input.** HappyHorse doesn't take video or audio references.
- **Editing is 1.0-only.** There's no 1.1 editing model.
- **Watermark in the API.** By default, "Happy Horse" is added in the bottom-right corner — you have to turn it off with a parameter.
- **Image requirements:** JPEG, PNG or WEBP, at least 400 px on the short side, up to 20 MB; small or heavily compressed sources give worse results.
- **Speech languages** aren't officially listed by Alibaba.

## Bottom line

HappyHorse is Alibaba's line of short cinematic clips with sound, with simple inputs: text, a frame or up to 9 images. Version 1.0 is still in Arena's top 10, and 1.1 holds references and motion better. Use HappyHorse for short dialogue scenes, product ads with a character and trailers. If you need clips up to 30 seconds or video and audio as input, go with [Wan 3.0](/models/wan-3-0); for 2K, [MiniMax H3](/models/minimax-h3). It's easy to compare them on the same prompt in AI2Face.

## FAQ

### What is HappyHorse?

Alibaba's AI video model and its happyhorse.com service. HappyHorse makes clips up to 15 seconds in 1080p with sound in one pass, from text, an image or a set of references.

### How is HappyHorse 1.1 different from 1.0?

According to Alibaba, 1.1 combines multiple references more accurately, handles complex motion more smoothly, understands prompts better, adds more detail and syncs sound with the picture more precisely. 1.1 also adds 480p. Video editing is still only available in the 1.0 model.

### How long are HappyHorse videos and what resolution?

From 3 to 15 seconds in 1-second steps, at 24 frames per second. HappyHorse 1.1 outputs 480p, 720p and 1080p; 1.0 outputs 720p and 1080p.

### Does HappyHorse have open weights?

No. Alibaba ships HappyHorse through its own service and cloud API. Sites calling the model open source have no backing in official materials.

### Which languages do HappyHorse characters speak?

Alibaba doesn't publish an official list of speech languages. Test speech in your language on a short 5-second clip before making a long one.

## Sources

- https://www.alibabacloud.com/blog/happyhorse-gets-stronger-motion-expressiveness-higher-generation-consistency-and-enhanced-visual-quality_603293
- https://www.alibabacloud.com/help/en/model-studio/video-generate-edit-model/
- https://www.alibabacloud.com/help/en/model-studio/happyhorse-reference-to-video-api-reference
- https://www.alibabacloud.com/help/en/model-studio/happyhorse-text-to-video-api-reference
- https://www.alibabacloud.com/help/en/model-studio/happyhorse-image-to-video-api-reference
- https://arena.ai/leaderboard/text-to-video