Cinematic Sequence
Post-trained variant · 480P and 768P · Synchronized audio
MiniMax H3 Max AI Video Generator
H3 Max ranks #1 for overall quality, prompt understanding, and aesthetics, and generates a five-second video in under three seconds.
Image to VideoReference to Video
Prompt *Describe your vision
Describe your vision. Type @ to reference uploaded media.
0/7000
Prompt Expansion Mode Balanced
One continuous take, no cuts
BLACK MIRROR TIDEFLATS / THE COLLAPSING SEA-CATHEDRAL
Five to fifteen seconds
480P or 768P at 24 FPS
Synchronized audio in one pass
See What MiniMax H3 Max Can Create
Explore six demonstrations with the full source prompt available on every card. Open any prompt to read or copy its complete direction.
Retro Gameplay
A side-scrolling city run
Spoken Scene
A direct-to-camera kitchen line
Clay Animation
A scarfed frog serves pancakes
Nature Macro
A hummingbird pauses at a flower
Product Close-up
Espresso forming in macro detail
What Is MiniMax H3 Max?
MiniMax H3 Max is a post-trained variant of the open-weight MiniMax H3 video model.
Additional training data targets aesthetics and prompt adherence, with infrastructure optimized for faster generation. H3 Max is not a separate official tier released by MiniMax: MiniMax produced the H3 base, while H3 Max is a specialized variant built for speed and prompt adherence.
The model is designed for rapid text-to-video, image-to-video, and reference-to-video work at 480P or 768P. Standard MiniMax H3 remains the appropriate option for 2K output and video editing.
MiniMax H3 Max Key Features
Each capability is paired with a demonstration that shows the behavior in a finished clip.
Camera & Motion
Direct the camera as precisely as the action
A low tracking shot can follow a scooter down a steep street while keeping the rider framed through the turn. The scene carries the requested speed into the background, body lean, camera path, and synchronized engine sound.
First & Last Frame
Connect two stills with continuous movement
Start with one image and optionally define the final frame. In this example, a balloon on damp morning grass rises into a wide aerial view, with the burner, wind, and upward motion carried through as one uninterrupted shot.
Character Consistency
Keep a character recognizable across cuts
The same person appears in six locations while the time, setting, and lighting change. Hair, clothing, facial features, and proportions remain coherent from a bright Victorian street to a sunset office scene within one generation.
Native Audio
Create the soundtrack in the same generation
Dialogue, ambience, foley, music, and environmental sound can be generated alongside the image. Describe the audio with the scene so timing and picture arrive together instead of requiring a separate synchronization pass.
Art Direction
Carry one visual language through the sequence
A defined art direction can persist while the scene changes. This 15-second example moves through six animated Greek-vase shots while retaining its palette, linework, gold accents, and hand-set lettering from beginning to end.
Prompt Adherence
Keep requested beats in their written order
H3 Max focuses its post-training on closer prompt adherence. Ordered actions are more likely to appear in sequence, while requested on-screen words are rendered more clearly. Specific directions give the model more structure to preserve.
MiniMax H3 Max vs MiniMax H3
Choose according to the workflow. H3 Max emphasizes fast 768P iteration, while standard H3 supports a broader toolset and higher resolution.
CapabilityH3 MaxStandard H3
Best fitSpeed and prompt adherenceResolution and broader control
Resolution480P / 768PUp to 2K
Text and image to videoYesYes
Reference to videoYesYes
Video editingNot a launch endpointYes
AudioSynchronizedSynchronized
How to Use MiniMax H3 Max
Four decisions take you from a first prompt to a reviewable clip. Keep each draft focused on one short sequence.
01
Pick text, image, or reference input
Start with text when the model should build the scene, upload opening and optional closing frames when the look is decided, or add image, video, and audio references to guide identity, motion, style, and sound.
02
Write the shot in order
List what happens beat by beat: subject, action, camera move, lighting, and sound. Put events in playback order and call out anything that must stay the same throughout.
03
Choose output settings
For a first draft, use 768P, five to ten seconds, and Balanced prompt expansion. Switch to 480P for faster iteration, or extend to fifteen seconds only when the scene needs the extra time.
04
Fix one issue per rerun
Check event order, camera path, on-screen text, character consistency, and audio sync. Change the weakest instruction once, then generate again before stacking more edits.
MiniMax H3 Max FAQ
Clear answers about the model's origin, output limits, reported speed, audio, and differences from standard H3.
What is MiniMax H3 Max?+
MiniMax H3 Max builds on the open-weight MiniMax H3 video model with additional post-training. The variant targets closer prompt adherence, stronger aesthetics, improved audio-visual output, and faster inference.
Is H3 Max an official model released by MiniMax?+
MiniMax developed the H3 base model. Additional post-training shaped the H3 Max variant and optimized the serving system for faster generation.
What resolution and duration does H3 Max support?+
Available outputs are 480P or 768P at 24 FPS, with clip lengths from five to fifteen seconds. Text-to-video and reference-to-video offer six aspect ratios, while image-to-video uses the proportions of the uploaded image.
Does MiniMax H3 Max support 2K?+
No. H3 Max is offered at 480P and 768P. Standard MiniMax H3 is the appropriate option when the workflow requires 2K output.
Can MiniMax H3 Max animate an image?+
Yes. Upload an opening image and describe how the scene should move. An optional final image can define where the generated shot should end.
Can MiniMax H3 Max use reference media?+
Yes. Reference-to-video accepts images, videos, and audio files together with a prompt. A request can include up to 12 assets: nine images, three videos, and three audio files.
How are H3 Max reference-to-video credits calculated?+
Output costs 1 credit per second at 480P or 2 credits per second at 768P. Each reference image adds 1 credit. Reference video adds 1 credit per second at 480P or 2 credits per second at 768P. Reference audio does not add a separate charge.
Does H3 Max generate audio?+
Yes. Dialogue, ambience, foley, music, and sound effects can be generated in sync with the video.
How fast is MiniMax H3 Max?+
Model inference is reported below three seconds for a five-second 768P clip. The complete wait may be longer because uploads, prompt expansion, queues, safety checks, and delivery happen outside inference.
MiniMax H3 Max or standard MiniMax H3?+
H3 Max suits fast 480P or 768P text-to-video, image-to-video, and reference-to-video generation with an emphasis on prompt adherence. Standard H3 remains the appropriate option for 2K output and video-editing workflows.
Get Started
Make Your First MiniMax H3 Max Video
Open the H3 Max generator, start from text or an image, and turn a five-to-fifteen-second brief into a 768P clip with synchronized sound.