June 23rd.AlibabaPublished yesterdayVideo Generation Model HappyHorse 1.1. Officially, the new version 1.0 has been updated in terms of dynamic performance, body consistency, command compliance, visual sense and audio。
On June 23rd, Ali Baba released video generation model HappyHorse 1.1 yesterday. Officially, the new version 1.0 has been updated in terms of dynamic performance, body consistency, command compliance, visual sense and audio. For the production specifications, the HappyHorse 1.1 single-time generation time of 3 to 15 seconds supports the 720p, 1080p resolution and free width ratio. The changes in capacity are concentrated in five areas:
Dynamic performance: Optimizing motor modelling and time-series consistency and improving the slowness and inefficiency of some of the images。
Main Consistency: Supports the simultaneous input of up to 9 role reference maps to reduce the drift of role and trade details for multi-play shorts, live live delivery, multi-person advertising, etc。
Directives are followed: reinforce context synonyms, scene planning and role modelling, and enhance the stability of camera programming under complex Prompt。
Visual perception: Reduces over-heavy, oil-skinned and smearing, and retains the details of pox, statute stripes, hair holes, etc。
Audio capacity: Station speed, pause and tone can be adapted to the scene and enhance understanding of sound descriptions such as background music, environmental sound。
At the same time, Ali announced that HappyHorse United Whale Recreation Group would host the "Horsepower" AI video contest, and that the winner would receive a million-dollar business list, with Zhang Ji-chul serving as a jury。