Page loading...
Page loading...
MiniMax H3, also known as Hailuo 3, is an open, general-purpose omni-modal video model that understands text, images, video, and audio together—and generates native-stereo 2K video up to 15 seconds.
MiniMax H3, also known as Hailuo 3 or Hailuo 03, is an open, general-purpose omni-modal generation model. It understands text, images, video, and audio together, then generates clips with native stereo audio at up to 2K resolution and 15 seconds in length. Raphael supports text-to-video, first-and-last-frame animation, and multimodal references. Output costs 48 credits per second; reference-video seconds and reference images after the fifth are charged separately.
See how H3 handles multimodal references, 2K detail, native stereo audio, and dynamic visual storytelling.
Use character, camera-motion, and ambient-sound references to create one coherent cinematic shot with synchronized sound and image.
H3's premise is not a collection of isolated modes: it puts creative intent and different reference materials into one shared context.
Before“Keep the reference character and café atmosphere, follow the camera's forward rhythm, and blend the surrounding sound naturally.”
H3 can understand text, images, video, and audio in one shared context. Give it your reference materials and describe how they should shape the shot, and it can interpret them as one creation task.
“A 2K product close-up: reveal material detail and packaging text with a slow orbit, then land on a clean final reveal.”
H3 supports whole-second clips from 4 to 15 seconds at a fixed 2K tier, using video representation and in-context regeneration to retain small text and visual detail.
“A night street tracking shot: foreground dialogue, distant traffic and music for spatial depth, with sound moving naturally as the camera turns.”
H3 jointly generates video and audio, including dialogue, ambience, music, and sound effects. Raphael keeps sound enabled because this model's outputs include native stereo audio.
“Turn this game menu into a 15-second animated interface: preserve the character, Start and Continue buttons, purple-and-lime palette, and graffiti texture while adding coherent motion to the background and accents.”
H3 works well for cinematic moments, animated game interfaces, and e-commerce ads. Describe the relationship between your reference material and the intended clip, and it can extend a still visual into continuous storytelling while preserving hierarchy and subject identity.
Switch between top video models in one generator.

Smooth motion and natural choreography.

Lower-cost drafts with rich reference control.

Cinematic 1080p video with synchronized native audio.

Fast, lower-cost text and image-to-video up to 1080p.

Native audio, 4K output, cinematic realism.

Fast action with bold camera moves.
Choose H3 for fixed 2K audiovisual output; choose Seedance when you want more resolution and cost options.
| Feature | MiniMax H3 | Seedance 2.0 |
|---|---|---|
| Resolution | Fixed 2K | 480p / 720p / 1080p |
| Duration | 4–15 seconds | 4–15 seconds |
| Native Audio | Always on, stereo | Optional |
| Reference Inputs | Image / Video / Audio | Image / Video / Audio |
| Standard Credits | 48 per second | 36 / 77 / 191 per second |
| Best For | Fixed 2K audiovisual finals | Flexible drafts through premium output |
Compare output goals, reference needs, and credits before choosing a model.
Free images include a watermark. Upgrade for clean outputs, faster generation, and commercial use.
Key facts about resolution, sound, duration, references, and credits.
Yes. MiniMax H3 is the product name used here; Hailuo 3 and Hailuo 03 are common aliases.
No. H3 produces native stereo audio with every video, so the sound control remains on.
Standard generation costs 48 credits per output second: 192 credits for 4 seconds, 240 for 5 seconds, 480 for 10 seconds, and 720 for 15 seconds.
Reference video input is added by its actual billable seconds. The first five reference images are included; each image from the sixth onward adds 16 credits.
Not currently. Audio references must be paired with an image or video.
Up to 9 images, 3 videos, and 3 audio clips, with no more than 12 files total. Reference videos and audio must be 2–15 seconds, and each media type can total at most 15 seconds.
H3 uses a fixed 2K quality tier and supports every whole-second duration from 4 through 15 seconds.
H3 offers one fixed 2K route with stereo audio always included. Seedance offers 480p, 720p, and 1080p tiers, including cheaper draft options and optional audio.