2K as Standard
Not a premium tier — every generation comes out at 2K.
Minimax H3 AI Video Generator
MiniMax's Hailuo 03 model produces silent 2K video, driven either by a first and last frame or by a set of reference images.
- Minimax H3
- 2K
데스크톱 브라우저에서 이 페이지를 열어 창작을 시작하세요.
The Highest-Resolution Video in the Roster, and Nothing to Configure
Minimax H3 is MiniMax's video model, and its position in Virse is unusually simple to describe: it produces 2K, it produces no sound, and there is nothing else to decide.
Most video models present a grid — resolution, aspect ratio, quality tier — and each choice is a small tax on getting started. This one has a single configuration. Every generation comes out at 2K with native audio, including lipsynced dialogue, so a clip arrives with sound rather than waiting on a scoring pass.
Use Minimax H3 when resolution is the requirement and audio is not. Produce detail-heavy product motion, character-led series driven by reference images, and footage destined for a timeline where the sound is being handled separately.
What Is Minimax H3?
Minimax H3 is an AI video generation model developed by MiniMax, known by its backend name Hailuo 03. Virse carries it as two entries with different inputs and the same output.
The model is built around three major strengths:
- 2K output as the single, standard resolution
- A reference-driven entry for holding a subject consistent across clips
- One configuration, with no audio switch and no quality tier to choose
Minimax H3 takes a first frame, a last frame, or both, plus a written brief. Minimax H3 Reference takes reference images instead, carrying their subjects, styling, and visual language into a generated sequence. Both output 2K with native audio.
Minimax H3 Specs at a Glance
Audio Always On
A generated audio track arrives with every clip, and there is no option to switch it off.
Two Entries
Frame-driven and reference-driven, listed separately.
One Configuration
No resolution selector and no quality grade. Audio is always on rather than optional.
Reference-Image Input
The Reference entry accepts supplied images as the driver.
First and Last Frame Control
The standard entry fixes either endpoint of the shot, or both.
Key Features of the Minimax H3 AI Video Generator
Resolution Without a Decision
2K is what this model does. There is no tier to weigh up and no upgrade path to consider, which removes the most common source of hesitation before a generation.
Sound by Design, Not by Option
Lipsync and ambience are produced in the same pass as the picture. That is the default here rather than a setting you remember to enable.
A Reference Entry for Continuity
When a character has to appear across six clips and stay recognisable, showing the model examples beats describing them. The Reference entry exists specifically for that.
Reference Roles You Assign
Several images can each carry a different job — one for the person, one for the setting, one for the styling — rather than being averaged into a single influence.
Predictable Endpoints
The standard entry accepts both a first and a last frame, which turns generation into a path between two known compositions instead of an open-ended one.
Identical Pricing Across Entries
Choosing between frame-driven and reference-driven is a question of what you have to work with, not of what you can afford.
Build with Minimax H3 in Virse
Reference-driven continuity work depends on the reference set staying stable. Change one image between clips and the character shifts, usually in a way nobody notices until the sequence is assembled. Virse keeps the set in one place. The reference images live on the canvas beside every clip generated from them, so the same inputs get reused rather than re-gathered, and drift becomes visible while there is still time to fix it.
Pin the Reference Set
Keep the character, product, and style images together on the canvas and point every clip at the same ones.
Generate the References First
Build the stills with an image model on the same canvas, then hand them to Minimax H3 without exporting.
Access 30+ Creative Models
Move between Minimax H3 and 30+ other image and video models without leaving the canvas or rewriting the brief.
Compare Clips in Sequence
Lay the run out in order so a subject that has drifted between shot two and shot five is obvious.
What Can You Create with Minimax H3?
Character-Led Series
Several clips featuring the same person, held together by a shared reference set.
Product Motion at 2K
Rotations, reveals, and detail passes at a resolution that survives cropping.
Fashion and Editorial
Garment and styling references driving movement without re-describing them each time.
Brand Sequences
Stylised clips where a defined visual language has to carry from shot to shot.
Location and Establishing Shots
Frame-driven clips where the composition is already decided.
Footage That Arrives With Sound
Material for timelines where generated dialogue and ambience are a starting point rather than something added later.
How to Use Minimax H3 in Virse
Pick the Entry
Frame-driven when you know the composition, reference-driven when a subject must stay recognisable.
Supply Frames or References
Load the two endpoint frames, or the reference set with a stated job for each picture.
Describe the Motion
State what moves in the scene, how the camera behaves, and at what pace.
Reuse the Setup for the Next Clip
Keep the same inputs and change only the brief so the series holds together.
How to Write a Minimax H3 Prompt
A useful Minimax H3 prompt usually includes four elements:
- Reference Roles
- What Moves
- Camera
- Ending
이렇게 쓰는 대신
A model walking through a gallery, cinematic, using the attached references.
이렇게 쓰세요
Use the person in Image 1, the coat in Image 2, and the gallery interior in Image 3. She walks from the left edge of frame toward the centre at an unhurried pace and stops facing a painting off camera to the right. The camera tracks with her at a constant distance, holding her at the same size in frame throughout. End with her stationary, in profile, at the centre of the frame.
Minimax H3 Prompt Examples
Reference-Driven Continuity
Use the woman in Image 1 and the workshop interior in Image 2. She is seated at a bench, hands working on something out of frame. She pauses, looks up toward the window on the left, then returns to her hands. Camera holds a static medium shot from slightly above bench height, no movement and no cuts. Preserve her face, hair, and clothing exactly as shown in Image 1. End with her head lowered again.
Product Detail Pass
Open on the composition in Image 1: a leather satchel standing upright on a pale stone ledge, buckles facing camera. The camera orbits slowly to the left through roughly forty-five degrees, keeping the satchel centred and at constant size. Light stays fixed as the camera moves, so highlights travel across the leather and hardware. End on a three-quarter view with the side gusset and stitching visible.
Frame-to-Frame Transition
Begin on Image 1: an empty stage with a single microphone stand, house lights up. The house lights fade down over the first half of the clip while a single overhead spotlight rises on the microphone. Nothing else moves. Camera holds a locked-off wide shot throughout. End on Image 2: the stage in darkness apart from the lit microphone.
Minimax H3 vs. Minimax H3 Reference
| 비교 항목 | Minimax H3 | Minimax H3 Reference |
|---|---|---|
| Input | First frame, last frame, or both | Supplied reference images |
| Output | Silent 2K | Silent 2K |
| Best for | Shots with a defined composition | Subjects that must stay recognisable |
| Endpoint control | Both ends can be fixed | Driven by references rather than frames |
| Typical use | Product motion, transitions | Character series, styled campaigns |
| Switching cost | Same brief structure on both | Same brief structure on both |
Tips for Better Minimax H3 Results
Assign Every Reference a Role
Uploading three pictures without saying which is the subject is the most common way these briefs go wrong.
Keep the Reference Set Stable Across a Series
Changing one image between clips is enough to break continuity. Change the brief instead.
Write Audio Direction
Output carries sound. Dialogue, effects, and ambience cues in the prompt shape what is generated.
Fix Both Frames for Transitions
Shots that move from one defined state to another are what the standard entry does best.
Minimax H3 FAQ
What is Minimax H3?
Minimax H3 is a video generation model from MiniMax, also known by its backend name Hailuo 03. Virse lists it as two entries — one frame-driven, one reference-driven — both producing 2K video with native audio.
What is the difference between Minimax H3 and Minimax H3 Reference?
The input. The standard entry takes a first and last frame; the Reference entry takes supplied reference images. Output is identical.
Is Minimax H3 free, and what does it cost?
Minimax H3 is available on Virse's paid plans. Generation is billed by the second from a monthly credit allowance, and the higher plans make it unmetered within fair use. Current plan rates are listed on the Virse pricing page.
Does Minimax H3 generate audio?
Yes. Audio is generated with the picture on every clip, including lipsynced dialogue, and there is no option to disable it.
What resolution does Minimax H3 produce?
2K, as its single output size. That is the highest resolution among the video models in Virse.
Can Minimax H3 keep a character consistent across clips?
Yes, through the Reference entry. Supply the same reference images for every clip in a series and change only the written brief.
Is there a quality tier or a resolution selector?
No. The model has one configuration, which is part of why it is quick to start with.
How do I use Minimax H3 in Virse?
Pick the entry matching your input, load the frames or references, write the motion brief, and generate.
One Rate, One Resolution, No Deliberation
Pick the entry that matches what you have, describe the motion, and get 2K back without a configuration screen in between.