Happy Horse AI Video Generator: 1.0 vs 1.1, Prices & How to Use

Happy Horse is Alibaba's AI video model with native audio. See what 1.0 and 1.1 can do, what they cost, and how to make Happy Horse videos on AI Studio.

A stylized glowing horse made of film frames galloping across a dark cinematic background with a subtle sound waveform in warm amber and deep blue tones
Table of contents

The Happy Horse AI video generator (also written HappyHorse) is an AI video model family built by the Future Life Lab team inside Alibaba Token Hub (ATH). It turns text prompts, a first-frame image, or up to nine reference images into 3 to 15 second 24 fps MP4 clips at up to 1080p with native audio, across current versions 1.0 and 1.1.

You can generate clips with Happy Horse 1.0 right in your browser on AI Studio by opening the Happy Horse video generator. AI Studio is an independent website, not affiliated with Alibaba, that runs several AI video and image models on a credit system and currently runs Happy Horse 1.0 (not 1.1).

What is Happy Horse AI and who made it?

Happy Horse AI is a closed-source video and audio generation model family created by the Future Life Lab team inside Alibaba Token Hub (ATH), Alibaba's AI innovation unit.

The model first caught attention around April 7, 2026, when HappyHorse-1.0 appeared anonymously on the Artificial Analysis Video Arena. In blind-vote rankings, it climbed to the number-one spot for both text-to-video and image-to-video in the no-audio categories. On April 10, 2026, Alibaba confirmed to CNBC that the anonymous model was its internal project.

Public API access for Happy Horse 1.0 followed in late April 2026, with fal listing availability from April 27, 2026 alongside Alibaba Cloud Model Studio. About two months later, on June 22–23, 2026, Alibaba released Happy Horse 1.1 with updated consistency, motion, aspect ratios, and lower 1080p pricing.

As for what happens under the hood, the team says Happy Horse uses roughly 15 billion parameters in a unified 40-layer self-attention Transformer that generates video and audio jointly in a single pass, though those architectural specs are team-reported rather than independently verified.

If you are looking for downloadable weights to run locally, be careful: Happy Horse is strictly closed source. There are no open weights, and fal explicitly notes that the model will not be open source or licensable. Any website offering "open source Happy Horse" downloads or local installers is not genuine. Also note an unrelated name clash: "Happy Horse 2" (happyhorse2.video) is an independent website that does not run Alibaba's model.

What can the Happy Horse AI video generator do?

The Happy Horse AI video generator creates 3 to 15 second 24 fps MP4 videos with synchronized sound from text prompts, a starting frame, or up to nine reference images.

Across its API endpoints, the model accepts prompts up to 5,000 non-Chinese characters (or 2,500 Chinese characters, with longer text truncated) and defaults to a 5-second clip length. Here is how each generation mode works.

Text-to-video

Text-to-video builds a scene from scratch using only your written prompt. You choose the duration (any integer from 3 to 15 seconds), the resolution (720p or 1080p, plus a 480p tier for 1.1 on Alibaba Cloud), and the aspect ratio. You can write a single continuous action or break the prompt into timed multi-shot instructions so the camera cuts between angles inside one clip.

Image-to-video

Image-to-video takes one starting image and animates it into a 3 to 15 second video. The input image serves as the first frame, and the output video automatically follows the aspect ratio of your uploaded picture. For clean results, the first-frame image needs to be at least 300 pixels on each side with an aspect ratio between 1:2.5 and 2.5:1.

Reference-to-video with up to 9 images

Reference-to-video lets you upload between 1 and 9 reference images to keep characters, objects, or settings consistent across a newly generated scene. Instead of just animating a starting frame, the model extracts the subjects from your uploaded pictures and places them into whatever shot and aspect ratio you choose.

On Happy Horse 1.0, ordering matters. You must refer to your uploaded images in exact order as character1, character2, character3, and so on in your prompt, where the first uploaded image is character1. If you leave out those explicit labels, the model may mix up your characters. Reference files should be JPEG, PNG, or WEBP with a short edge of at least 400 pixels; clear 720p or higher images with proportions close to your target video aspect ratio work best.

Native audio and lip-sync

Unlike video models that output silent clips or bolt on sound effects in a separate post-processing step, Happy Horse generates audio in the same pass as the picture. When you describe a scene, it produces matching ambience, sound effects, and spoken dialogue automatically. If you include spoken lines in quotation marks inside your prompt, the model generates multilingual lip-sync timed to the character's mouth movements.

Video edit (API only, 1.0 only)

At the API level, Happy Horse 1.0 also includes a dedicated video editing endpoint (happyhorse-1.0-video-edit) designed to restyle or replace elements inside an existing video clip. This mode was not updated for 1.1 and remains a 1.0-only API endpoint. AI Studio does not offer the video edit mode.

Happy Horse 1.0 vs 1.1: what changed?

Happy Horse 1.1 improves character and product consistency across multiple references, smooths out fast action motion, tightens prompt adherence and audio-video sync, adds four new aspect ratios, and lowers the price of 1080p generation compared to 1.0.

Here is a side-by-side look at how the two versions compare across API providers and on AI Studio:

FeatureHappy Horse 1.0Happy Horse 1.1
Release dateApril 2026 (Arena ~April 7; API late April)June 22–23, 2026
ModesText-to-video, image-to-video, reference-to-video (1–9 images), video editText-to-video, image-to-video, reference-to-video (1–9 images)
Aspect ratios5 ratios: 16:9, 9:16, 1:1, 4:3, 3:4 (image-to-video follows input)9 ratios: adds 21:9, 9:21, 5:4, 4:5 (image-to-video follows input)
Resolutions720p, 1080p720p, 1080p (plus 480p on Alibaba Cloud)
fal API price (720p)$0.14 per second$0.14 per second
fal API price (1080p)$0.28 per second$0.18 per second
Native audioYes (joint audio-video pass with lip-sync)Yes (with improved audio-video sync)
Available on AI StudioYesNot yet

Happy Horse 1.0 is still a sensible choice for everyday clips. Its 720p rate matches 1.1 on fal, its five aspect ratios cover landscape, vertical and square formats, and it keeps the same 3 to 15 second range, native audio and up to nine reference images. If you don't need ultrawide 21:9 framing or the cheaper 1080p rate, 1.0 covers most short social and product clips, and it is the version you can run in the browser on AI Studio.

Is there a Happy Horse 2.0?

As of November 13, 2026, there is no Happy Horse 2.0 release, model ID, or pricing available.

At a Venice Film Festival panel in early September 2026, Alibaba Token Hub stated that Happy Horse 2.0 and a companion tool called "HappyHorse Control" were planned for the end of September, but that release did not ship on schedule. Later at its September 2026 conference, Alibaba also said a next-generation video model would arrive in November. Until an actual model ID and endpoint go live, treat any "Happy Horse 2.0" spec sheets floating around online as unconfirmed.

How much does Happy Horse cost?

Happy Horse pricing is calculated per second of generated video, ranging from $0.14 to $0.28 per second on developer APIs and 10 to 18 credits per second on AI Studio.

If you connect directly through a cloud API, exact per-second prices vary by provider and region. As listed in mid-to-late 2026:

  • fal (Happy Horse 1.0): $0.14 per second at 720p, and $0.28 per second at 1080p.
  • fal (Happy Horse 1.1): $0.14 per second at 720p, and $0.18 per second at 1080p. For example, a 10-second 1080p video costs $2.80 on 1.0 versus $1.80 on 1.1.
  • Alibaba Cloud Model Studio (international deployment list price for happyhorse-1.0): $0.14 per second at 720p, and $0.24 per second at 1080p.

If you prefer generating in the browser without setting up API keys, AI Studio runs Happy Horse 1.0 on a straightforward per-second credit formula. Total cost equals the per-second rate multiplied by the number of seconds, rounded up:

ResolutionPer Second5 Seconds (Default)10 Seconds15 Seconds (Max)
720p10 credits50 credits100 credits150 credits
1080p18 credits90 credits180 credits270 credits

A minimum 3-second clip at 720p costs 30 credits. New AI Studio accounts receive 4 trial credits upon sign-up, which lets you look around the interface, though you will need more credits to run a Happy Horse generation. AI Studio shows the exact credit cost for your chosen duration and resolution before you click generate.

How to make a Happy Horse video on AI Studio

You can make a Happy Horse video on AI Studio in seven quick steps right in your web browser without managing code or cloud billing.

  1. Open the generator: Go to the dedicated Happy Horse page on AI Studio, or open the main AI video generator and select Happy Horse 1.0 from the model selector (alongside other browser models on the site like Seedance, Kling, Gemini Omni, MiniMax H3, and Grok Imagine).
  2. Choose your generation mode: Pick Text to Video, Image to Video, or Reference to Video depending on whether you are starting from a blank slate, a first frame, or character reference pictures.
  3. Upload your image(s) if needed:
    • For Image to Video, upload one image (at least 300 px per side, ratio between 1:2.5 and 2.5:1) to serve as the opening frame.
    • For Reference to Video, upload 1 to 9 clear JPEG, PNG, or WEBP images (short edge at least 400 px, 720p+ recommended) in the exact order you plan to reference them.
  4. Write your prompt: Enter your scene description, camera motion, and any spoken dialogue or sound effects (up to 5,000 non-Chinese characters or 2,500 Chinese characters). If you are using Reference to Video, label your uploaded subjects in order as character1, character2, and so on.
  5. Set duration, resolution, and aspect ratio:
    • Pick any whole-second duration from 3 to 15 seconds (5 seconds is the default).
    • Select 720p (default) or 1080p.
    • For Text to Video and Reference to Video, choose an aspect ratio: 16:9 (default), 9:16, 1:1, 4:3, or 3:4. If you are in Image to Video mode, aspect ratio controls are hidden because the output automatically matches your uploaded image's dimensions.
  6. Check the credit cost: Look at the credit readout before submitting—for instance, 50 credits for a 5-second 720p clip or 90 credits for 5 seconds at 1080p.
  7. Generate and download: Click generate, wait for the single-pass audio and video render to finish, and download your MP4 clip.

What AI Studio doesn't offer for Happy Horse

To save you time before you buy credits, here is what AI Studio currently does not support for Happy Horse:

  • Happy Horse 1.1: AI Studio runs 1.0 only, so 1.1 and its four extra aspect ratios (21:9, 9:21, 5:4, 4:5) are not available here yet.
  • Video Edit mode: You cannot upload an existing video to restyle or replace objects (happyhorse-1.0-video-edit is API-only).
  • 480p resolution: Only 720p and 1080p are offered.
  • Seed control and watermark toggles: Custom seed numbers and watermark toggles are not exposed in the interface.
  • Video or audio file inputs: You can only input text and still images (no driving audio tracks or input video files).
  • Out-of-range durations: Clips under 3 seconds or over 15 seconds are not supported.

Prompt tips for Happy Horse

Writing structured prompts for the Happy Horse AI video generator helps you take full advantage of its multi-shot cutting, reference tracking, and native audio engine.

  • Follow the Scene + Subject + Motion + Audio formula: EvoLink's Happy Horse documentation recommends organizing prompts in four clear parts. Establish the environment and lighting first, describe who or what is in frame, specify how the subject and camera move, and finish by stating the background ambience, sound effects, and spoken lines.
  • Use shot-by-shot time ranges for multi-cut clips: Because Happy Horse supports clips up to 15 seconds, a single static shot can feel slow. You can direct camera cuts inside one generation by writing explicit time brackets and shot scales—such as Shot 1 [0~3s] wide angle: ...; Shot 2 [3~6s] medium shot: ...—so the model knows exactly when to switch angles.
  • Label reference images as character1, character2 in order: In Reference-to-Video mode, never rely on vague descriptions like "the woman in the red jacket" alone. Always tie each uploaded file to its slot number (character1 for the first image, character2 for the second) and keep your uploaded image proportions close to your target output ratio.
  • For image-to-video, describe the motion rather than re-describing the picture: The model can already see the colors, clothes, and layout of your first-frame image. Spend your prompt words explaining what changes after frame one: how the subject turns, how the wind moves the background, where the camera pans, and what sounds play.
  • Keep spoken dialogue short enough to fit the duration: Native lip-sync works best when the line can be spoken naturally inside the shot's time window. A 3-second shot only has room for a short sentence, whereas a 10-second clip gives characters time to exchange two brief lines without rushing the mouth animation.

Here is an example of a text-to-video prompt using timed shot cuts and native audio:

Shot 1 [0~4s] wide angle: A rainy neon-lit noodle stall in Tokyo at night, steam rising from a large copper pot. A chef in a dark apron ladles broth into a ceramic bowl as rain taps against the canvas awning.
Shot 2 [4~8s] medium close-up: The camera pushes in gently as the chef sets the bowl on the wooden counter, looks at the camera with a warm smile, and says in English, "Careful, the broth is right off the stove." Audio: steady rain on canvas, bubbling soup, soft ceramic clink on wood, and clear spoken voice.

And here is an example of a reference-to-video prompt using two uploaded character images:

Shot 1 [0~5s] medium two-shot: Inside a sunlit brutalist concrete art gallery, character1 and character2 stand side by side looking at a tall abstract brass sculpture. character1 gestures toward the top of the sculpture and says, "The balance on that upper piece looks impossible."
Shot 2 [5~10s] over-the-shoulder shot: Focusing on character2 nodding thoughtfully while taking a slow step closer to inspect the metalwork, camera tracking slightly right. Audio: quiet museum room tone, soft footsteps on polished concrete, and natural indoor dialogue with subtle echo.

Whether you want to animate a single illustration with synchronized sound effects or stage a 15-second multi-shot dialogue scene using reference photos, Happy Horse 1.0 puts video and audio generation into a single step. Pick a prompt structure, line up your reference images if you have them, and try generating your first clip in the browser on AI Studio.

Sources

Frequently asked questions

Who made Happy Horse AI?

Happy Horse (also written HappyHorse) is an AI video generation model family from Alibaba. It was built by the Future Life Lab team inside Alibaba Token Hub (ATH), which is Alibaba's AI innovation unit. After the model debuted anonymously on the Artificial Analysis Video Arena around April 7, 2026, Alibaba confirmed on April 10, 2026 that Happy Horse was its project.

What is the difference between Happy Horse 1.0 and 1.1?

Happy Horse 1.1 released on June 22–23, 2026 with better multi-reference consistency for characters and products, smoother action motion, stronger prompt adherence, and tighter audio-video sync. It also adds four aspect ratios (21:9, 9:21, 5:4, and 4:5) for nine total, adds a 480p tier on Alibaba Cloud, and lowers 1080p API pricing. However, video editing remains a 1.0-only endpoint.

Is Happy Horse open source?

No, Happy Horse is closed source and does not have open weights. According to fal, the model will not be open source or licensable. Any websites or repositories offering downloadable open-source Happy Horse weights are not genuine. You can only access the model through cloud APIs like Alibaba Cloud Model Studio and fal, or through browser platforms like AI Studio.

How much does Happy Horse cost?

API pricing is billed per second of output and varies by provider and region. On fal, Happy Horse 1.0 costs $0.14 per second at 720p and $0.28 per second at 1080p, while 1.1 drops 1080p to $0.18 per second. On AI Studio, Happy Horse 1.0 costs 10 credits per second at 720p and 18 credits per second at 1080p.

Is there a Happy Horse 2.0?

As of November 13, 2026, there is no Happy Horse 2.0 model ID, pricing, or release. Alibaba Token Hub mentioned at a Venice Film Festival panel in early September 2026 that 2.0 and a HappyHorse Control tool were planned for late September, and Alibaba promised a next-generation video model for November, but it was trailed rather than shipped.

Create with AI Video Studio

Turn your idea into a video

Explore the video generator and start creating.

Generate a video

Written by

AI Studio Team