HeyGen Video 1

HeyGen's first general-purpose video model, from the company best known for talking avatars and digital twins. Instead of animating a presenter, it generates the whole scene from one prompt: the subject, the setting, the lighting and the matching sound, dialogue, ambience and effects together, in clips of 5 to 30 seconds at up to 2K, in styles from live action to claymation, anime and screen print. It takes text, images and references, and runs in HeyGen's app, a free web playground and the API.

It is built on MiniMax H3 and post-trained by HeyGen, a sign of where the open release of a top video model led within two months: an application company could take the base, train it on its own, and ship a model of its own. In a blind test on Artificial Analysis it ranked first in text-to-video without audio, ahead of Wan 3.0, H3 itself and Seedance 2.5, and third once sound was counted. Late in October HeyGen extended single takes to 30 seconds and cut the API price to a cent a second.