Video model · from 78 credits
Generate videos with MiniMax H3
MiniMax H3 is the third Hailuo generation, known for the liveliness of its camera moves. It accepts up to nine reference images and an end frame, which makes it a true directed-shot tool: give it the start point, the destination and the character references, and it builds the journey. Clips of five to ten seconds, audio available. Its per-second cost sits mid-range, justified as soon as the camera has to move. It also reads reference voices: name your audios "Audio 1, Audio 2" in the prompt, alongside at least one image, and the character speaks with the provided voice, lips synchronized.
Capabilities
- Up to 9 reference images
- Reference voices: up to 5 wired audios, the character speaks with the provided voice
- End image (first → last frame interpolation)
- Audio generated with the video
- Durations: 5–10 s
- Aspect ratios: 21:9, 16:9, 4:3, 1:1, 3:4, 9:16
A real directed-shot tool
MiniMax H3 is the third Hailuo generation, known for how lively its camera moves are. What sets it apart in the catalogue is how many inputs it takes: up to nine reference images, plus an end frame. You give it the starting point, the arrival point and the character references, and it builds the path between them.
That is what separates it from a model freely interpreting a sentence. When a shot has to land on a precise image, a product in close-up, a face found again, a composition to respect, the end frame changes everything: the movement is constrained, so it is usable in an edit.
Five to ten seconds, and why not more
The model produces clips of five to ten seconds. The ceiling is not the model's limit but the service's: the gateway only exposes the synchronous call, with no way to resume if the compute runs past its allotted time. Beyond ten seconds, generation only landed once in three, and the failures were refunded but lost in time.
Ten seconds is enough for one shot. For a longer sequence, the method is to chain two clips using the last frame of the first as the starting image of the second, which the canvas allows by wiring the nodes one behind the other.
The reference voice
H3 also accepts voices as input. Name your audio files in the prompt, Audio 1, Audio 2, attach at least one image, and the character speaks with the supplied voice, lips in sync. Real faces pass, which is not the case for every video model in the catalogue.
That combination is what makes the model interesting for an avatar or a testimonial: an image of the character, a voice recorded or cloned elsewhere, and a ten-second shot where the two hold together. Its per-second cost sits mid-range, justified as soon as the camera has to move.
MiniMax H3 next to its closest siblings
Figures pulled from the same registry as billing: when a price or a cap changes, this table changes with it.
| MiniMax H3 | Veo 3.1 Fast | Kling V3 | |
|---|---|---|---|
| Base price | 78 credits / 5s | 72 credits / 4s | 91 credits / 3s |
| Max duration | 10s | 8s | 15s |
| Max resolution | – | 2160p | 2160p |
| Generated audio | Yes | Yes | Yes |
| Reference images | 9 | 3 | 3 |
| Reference voices | 5 | No | No |
How to use it on Imaginode
- 1Validate your scene as an image first (1 to 7 credits per try), then wire it into a Video node.
- 2Pick MiniMax H3, a duration and a resolution: the exact cost shows before you click.
- 3Describe the motion in the prompt, add a Camera node to direct the shot, and generate.
Frequently asked questions
What is the maximum clip length on MiniMax H3?
Ten seconds. That ceiling comes from the service, not the model: beyond it, generation did not land often enough to be offered honestly. For a longer sequence, chain two clips.
Does it accept a real face as a reference?
Yes, unlike several video models in the catalogue that refuse real faces. That is what makes it usable for an avatar or a testimonial.
How do I make it speak with a specific voice?
Attach the audio file and name it in the prompt, Audio 1 for instance, together with at least one image of the character. The model syncs the lips to the supplied voice.
How much does a video generated with MiniMax H3 cost?
78 credits for 5s at 720p, about $0.94 (1 credit = $0.012). The exact price for each duration and resolution shows on the Generate button before you launch, and failed generations are refunded automatically.
Which durations and resolutions does MiniMax H3 offer?
Duration adjustable from 5 to 10s. Soundtrack generated along with the picture at no extra cost: the rate is the same with the sound on or off. Up to 9 reference images to keep a character or set consistent. End image accepted to control the final frame. Up to 5 reference voices wired from the canvas: the character speaks with the provided voice, lips synchronized.
Can a character speak with a specific voice on MiniMax H3?
Yes: wire the output of a Voice-over node (catalog voice or cloned voice) to the video node's Voices port, and MiniMax H3 reuses the vocal take as is, words, timbre and pace, lips synchronized. The exact price shows on the Generate button before you launch.
Can I try MiniMax H3 for free?
Yes: signing up grants trial credits, no credit card required. Enough to test MiniMax H3 in real conditions before deciding on a plan or a top-up.
Do I need to install anything to use MiniMax H3?
No: MiniMax H3 runs server-side, in the browser, mobile included. Imaginode's node canvas wires its renders to the other models (references, camera, styles) with nothing to install.
MiniMax H3 at the competition
MiniMax H3 or its variants are also available at Hailuo AI (MiniMax). Pricing grid recorded on a given date, sources cited: the comparison details what credit-based billing changes.
Read the Imaginode vs Hailuo AI (MiniMax) comparison →
Try MiniMax H3 on a real canvas
105 models, one account, credits that work everywhere. The price always shows before you generate.
Get started
Grok Imagine Video
Kling 2.5 Turbo
Wan 2.7