Alibaba officially launched its multimodal video generation model HappyHorse 1.0 (internal codename “Happy Horse”) in late April 2026, currently in a gray-scale testing phase. The model has topped multiple sub-leaderboards on the Artificial Analysis Video Arena, becoming one of the most watched newcomers in the video generation space.
Core Specifications
HappyHorse 1.0 supports four modes: text-to-video, image-to-video, video editing, and reference-to-video. A single generation produces 1080P resolution video ranging from 3 to 15 seconds, with a full generation cycle of approximately 2 to 5 minutes. The model can process prompts up to 800 words in length and supports multi-shot narrative structures.
Audio-video joint generation is a core feature — the model synchronously produces voiceover and ambient sound alongside the visual output, eliminating the need for separate post-processing. Lip sync functionality covers seven languages: Chinese, English, Japanese, French, German, Spanish, and Arabic.
Leaderboard Performance
On Artificial Analysis Video Arena’s image-to-video (with audio) sub-leaderboard, HappyHorse 1.0 ranks first, surpassing Seedance 2.0 which previously held the top spot for an extended period. In the text-to-video and video editing sub-categories, the model also enters the top three.
Multiple third-party platforms have integrated the model, including Venice, OpenArt, APIMart, Muvi AI, Renoise, Pollo AI, and HIX AI. Several platforms are offering limited-time discounts during the initial launch period.
Community Feedback
Early gray-scale testing users widely praised the model’s performance in portrait close-up scenarios. In portrait generation at 35mm to 85mm focal lengths, the background bokeh effect and character detail preservation have received positive reviews. Multiple users engaged in overseas short drama production noted that the naturalness of generated faces has significantly improved compared to previous products, making it suitable for direct use in commercial content production.
However, some users reported that when characters are placed in large-scale scenes, the model occasionally exhibits overfitting. The model still has room for optimization in complex scene composition.
Pricing and Availability
HappyHorse 1.0 is currently in gray-scale testing, with some platforms lowering the barrier to entry by offering free credits. Official pricing has not yet been announced; existing API service providers charge approximately 90 credits per generation.
During the gray-scale testing period, model access may be adjusted at any time. Users are advised to follow official platform announcements for the latest information.