HappyHorse 1.0

HappyHorse 1.0 is a video generation model from Alibaba, built by its ATH unit (the Taotian Future Life Lab, led by Zhang Di, who previously headed Kling at Kuaishou). It first appeared anonymously on the Artificial Analysis Video Arena in early April 2026 and climbed to number one in both text-to-video and image-to-video blind preference voting before Alibaba confirmed it was theirs. It entered limited beta on April 28, 2026.

According to the developers and their API partners, the model uses a unified single-stream Transformer with 40 self-attention layers that processes text, image, video, and audio tokens in one sequence. Video and soundtrack are generated together in a single pass rather than having audio added afterward, which enables lip-synced dialogue in seven languages: Mandarin, Cantonese, English, Japanese, Korean, German, and French. These architecture details come from developer materials and are not independently confirmed.

The model covers text-to-video, image-to-video and subject-to-video generation, plus editing modes that modify an existing video or swap its subject while keeping motion and composition. Clips run 3 to 15 seconds at up to 1080p with multi-shot sequencing and synchronized audio. HappyHorse is closed source.

Alibaba positions the model for advertising, e-commerce, short-form and social content, and the team has said the project is still under development. In June 2026 Alibaba followed it with HappyHorse 1.1, which it says reads multiple reference images better and moves more smoothly; this page runs 1.0.

On Vidney, HappyHorse 1.0 runs text to video and image to video at 720p or 1080p, any whole second from 3 to 15, and every clip comes back with sound. Text to video takes 16:9 or 9:16; image to video keeps the shape of the picture. The model's own mark is switched off, and downloads on the free plan carry a small Vidney watermark instead. Subject to video and the editing modes are not offered here.