April 9th news, yesterday, a name HappyHorseIT'S ANONYMOUSVideo Generation ModelA sudden appearance on the Video Arena list of the authoritative AI assessment platform, Artificial Analysis, and the overwhelming appearance of many tracks。
(a) In the text-repeated video (no audio) track, HappyHorse-1.0 is first in the Elo score of 1332, leading by the second name of Dreamina Seedance 2.0 nearly 60 minutes
(a) In the video-to-video (no audio) track, it ran a higher score of 1391, updating the historical record of the list
Even the audio lanes, which require a high degree of audio-visual synergy, are the second most stable in the worldNext to Seedance 2.0.
HappyHorse-1.0 is published in a very low-key manner — no product launch, no technical blog, and no business endorsement. Its official network has happyhorse.app provides two modes of text- and image-to-video support for the original output of 1080p and emphasizes multi-photo narrative consistency and flow-on synthesis for free registration。
Netizen Vigo Zhao (@VigoCreativeAI) wrote on X that after comparing HappyHorse-1.0's open baseline data with known models, it was found to correspond to the daVinci-MagiHuman, an open source in GitHub in March this year:
Many data, such as visual quality, text alignment, physical consistency, voice-charter error rates, are almost identical, as are the presentation styles of the official network architecture, performance tables and presentation videos。
daVinci-MagiHuman was officially opened on March 23 of this year and was developed jointly by the Shanghai Innovation Institute (SII) Generating Artificial Intelligence Research Laboratory (GAIR) and Sand.ai, Beijing。