MiniMax H3, also known as Hailuo 03, is a multimodal video model that can work from text, images, video, and audio references and can generate native-audio video.
This page offers text-to-video, first-and-last-frame image-to-video, and reference-to-video with up to nine images plus one optional video and one optional audio file in the current interface.
The current integration supports output durations from 4 through 15 seconds.
You can select 768P or 2K. The displayed credit estimate changes with resolution and duration.
Yes. H3 can generate native audio directed by the prompt, including dialogue, music, ambience, and sound effects, although generated audio should still be reviewed for timing and quality.
Yes. In Frames mode, upload a required first frame and an optional last frame, then describe the motion and transition between them.
Output is billed per generated second. If Reference mode includes a video, input-video seconds are also billed; the first five input images are included and each additional image adds credits.
Assign a clear role to every asset, such as identity from an image, movement from a video, and voice tone from audio, then state which subjects, designs, and scene details must remain unchanged.
MiniMax H3, also known as Hailuo 03, combines text, first-and-last-frame, and multimodal reference workflows for native-audio video generation.
No public result is stored for this Hub yet. Use the examples below to create the first model-specific result.
Example prompt
A compact silver speaker rotates on a black glass pedestal. Move from a macro grille detail to a wide hero shot, synchronize each light pulse with a deep electronic beat, preserve the logo and product proportions, premium studio commercial.
Example prompt
Transition naturally from the empty daytime storefront in the first frame to the illuminated evening launch event in the last frame. Add guests gradually, keep the architecture and signage fixed, and build the ambient crowd sound with the scene.
Example prompt
Use the portrait for character identity, the reference video for restrained body movement and camera rhythm, and the audio reference for vocal tone. Keep the face, jacket, and background palette consistent across the full sequence.
When several assets are supplied, state what each image, video, or audio file should control to reduce conflicts and drift.
Reference-video seconds are billed with output seconds, and images beyond the first five add input-image credits. Review the displayed estimate before generating.
Faces, hands, small text, logos, timing, audio synchronization, and unchanged areas still need human review before publishing.