Recently came across this project. Artificial Analysis currently lists HappyHorse-1.0 at #1 on the Text-to-Video (No Audio) leaderboard, and their Image-to-Video page also indicates it leads there as well.
The leaderboard ranking is from Artificial Analysis; the architecture details on the site are from the project’s own public materials.
Interesting mainly because it seems to combine strong blind-preference results with a fairly ambitious public technical story around a unified multimodal video model.