On June 22, 2026, Alibaba officially released the video-generation model HappyHorse 1.1. As an upgraded version of the HappyHorse series, the model achieves systematic optimizations across key dimensions—including dynamic expressiveness, subject consistency, instruction following, visual quality, and audio capabilities—resulting in a significant overall performance improvement. HappyHorse 1.1 is now available via API on the Alibaba Bailian platform, with plans to open more features in the future.

Alibaba says HappyHorse 1.1 builds on the 1.0 version’s strengths in joint audio-video generation and iterates based on user feedback and real-world use cases, aiming to provide content creators, marketing teams, and film producers with a more professional and user-friendly AI video-generation tool.

Key Upgrades in HappyHorse 1.1

1. Significant improvements in dynamic expressiveness and motion naturalness HappyHorse 1.1 achieves breakthroughs in motion smoothness and physical realism. The model better understands complex movements (such as walking, object interactions, and natural phenomena), greatly improving dynamic coherence in generated videos. Official tests show that in blind tests of dynamic scenes, version 1.1’s win rate improved by about 25%–30% over 1.0.

2. Improved subject consistency The model’s ability to preserve the appearance, clothing, and features of subjects (people, animals, objects) has been significantly enhanced. Even in long-sequence videos, subjects are less likely to exhibit degradation or deformation. This improvement is particularly suitable for narrative videos and advertisements that require stable character imaging.

3. Enhanced instruction-following HappyHorse 1.1 has stronger comprehension and execution capabilities for complex text prompts, supporting finer control of camera movement, specified styles, and scene descriptions. Users can precisely control camera motion, composition, and atmosphere through natural language.

4. Visual quality and detail rendering Image clarity, color reproduction, and lighting effects have been optimized—especially at 1080p resolution, where texture detail and cinematic feel are stronger. The model supports 720p and 1080p output and can flexibly adjust aspect ratios.

5. Upgraded audio capabilities As a model that natively supports joint audio-video generation, HappyHorse 1.1 further improves lip-sync, environmental sound effects, and background music matching, supports multi-language lip synchronization, and reduces word error rate.

Technical specifications:

Single generation duration: 3–15 seconds
Resolution: 720p / 1080p
Supports text-to-video, image-to-video, reference-based generation and other modes
Architecture: efficient Transformer structure with notable inference speed optimizations

Performance metrics

According to Alibaba’s official tests and third-party platforms (such as Artificial Analysis), HappyHorse 1.1 performs well across multiple video-generation benchmarks:

In the Artificial Analysis video arena, its overall Elo score remains leading.
Blind-test win rates for dynamic expressiveness and subject consistency have improved significantly compared with version 1.0.
Audio synchronization accuracy and visual quality scores have both shown clear improvements.

Actual user feedback indicates that HappyHorse 1.1 is practical for scenarios such as advertising short videos, short drama trailers, educational content, and social media materials, with generation quality approaching professional film standards.</