You write a scene where a character says a line, the voice sounds right, and then the video plays it over a face that never moves, or a mouth that flaps out of time with the words. That mismatch is the fastest way to make an AI video feel fake. EasyVid's Classic mode now has a lip-sync model for exactly those scenes: MiniMax H3 Max (Lip Sync).
What MiniMax H3 Max is
H3 Max is fal's version of MiniMax's H3 video model. fal describes it as "a post-trained variant of MiniMax H3, tuned for stronger prompt adherence and better aesthetics." Its reference-to-video mode takes reference images, video and audio, and can open on a set first frame and end on a set last frame.
How Classic uses it
In Classic, every scene starts as an image you generate and approve. The lip-sync model keeps that picture: the clip opens exactly on your scene image, can end on an end frame you choose, and the speaking character's lips follow the voice audio of the scene's lines. The clip keeps that audio, so the words and the mouth stay together in the final video.
You don't have to pick it scene by scene. With the animation model set to Auto, EasyVid uses MiniMax H3 Max (Lip Sync) for scenes where a character speaks on screen and MiniMax H3 Max Turbo everywhere else, including narrated scenes, where nobody on screen should be talking.
What it costs
MiniMax H3 Max (Lip Sync) generates at 768p, runs as long as the scene's audio (5 to 15 seconds), and costs 15 credits per second. Scenes without on-screen dialogue stay on MiniMax H3 Max Turbo at 5 credits per second.
Getting good lip sync
- Show the speaker's face. The scene image should show the person who talks, facing the camera or in three-quarter view, large enough to read their mouth.
- One speaker per shot works best. When two characters talk, a conversation cut between them reads more clearly than both talking in one frame.
- Generate the voices first. The clip follows the scene's line audio, so make the voices sound right before you animate.
Lip sync in Director mode
Director mode also uses MiniMax H3 Max, alongside Seedance 2.5, for scenes built from character and location references rather than a set first frame. If you want to mix models within one film, each scene can use the one that suits it. See how characters talk on camera in Director mode.