MiniMax H3 (Hailuo 3.0)
Generate 2K video at 24fps with dialogue, effects, and ambience mixed in the same pass. Combine image, video, and audio references in one prompt, then cut between shots without stitching takes together. Try MiniMax H3 on MaxH3 for free.
Key Features of MiniMax H3
- Multimodal Understanding: Describe how your references relate and H3 works out the rest.
- Native Stereo Audio: Dialogue, effects, and ambience rendered in the same pass as the frames.
- Omni Reference: Images, clips and audio as references in one generation — nine, three and three.
- Native Multi-Shot: Cut between wide, medium, and close-up without stitching takes together.
- Camera Control: Write the move you want, or lift it from a reference clip.
- First & Last Frame: Set both ends of the shot and H3 solves the motion between them.
- Precise Video Editing: Swap subjects, objects, lighting, and dialogue by naming the change.
Multimodal understanding across every input
H3 does not read your references one at a time. Hand it a clip, a photo, and a voice recording, then say how they connect: take the camera move from the first, put the person from the second in front of it, match the delivery to the third. The model works out which file supplies what and how the pieces combine, so direction stays in plain language instead of a stack of separate uploads and settings. You describe the shot the way you would describe it to a crew, and H3 resolves the rest.
INPUT
Reference image
Reference clip
Reference audio
OUTPUT VIDEO
0:06Native stereo audio in the same pass
Most models hand you a silent clip and leave the sound to a second tool. H3 renders dialogue, effects, room tone, and score alongside the frames and delivers them as native stereo in one file. Timing is set against what happens on screen, so a door closes when it closes and footsteps land where the feet land. Nothing to sync afterwards, and nothing to rebuild when the shot changes.
Omni Reference takes images, clips and audio together
Real briefs are rarely one input, and rarely one kind of input. Omni Reference accepts all three in the same generation — up to nine images, three video clips and three audio files, each capped on its own. Lock a face from a photo, pull motion from footage, bring a recording along with it, and carry the lot into the same shot. Audio references travel with an image or a clip rather than on their own. It is the same idea behind reference to video, with far more room to work.
INPUT
OUTPUT VIDEO
0:10Native multi-shot, no stitching
Fifteen seconds is long enough for a sequence, not just a moment. H3 cuts between shots inside one generation: a wide establishing frame to a medium tracking shot to a close-up, with the character and setting holding across all three. Write the beats in order and the model handles the transitions. One render, one audio bed, nothing to assemble afterwards.
Camera control without a preset menu
H3 reads camera language as language. Ask for a slow push-in, a crane sweep, or a handheld follow and it holds, no mode to select and no parameter to tune. Point it at a reference clip instead and it lifts the movement, the timing, and the camera behaviour onto a completely different subject. Useful when you already have footage that moves the way you want but a subject that does not match.
First and last frame, motion in between
Give H3 a single image and it generates forward from that frame. Give it two and it works out how to get from one to the other: the transition, the pacing, the physics of the move. When you know how a shot opens and how it lands but not what happens in the middle, that middle is exactly what H3 is solving for.
Precise video editing by description
Generation is only half of it. Replace a character, add or remove an object, change the background, relight the scene, rewrite the dialogue, or change whose voice delivers it. The change applies to what you named and leaves the rest of the frame alone, so you revise a take instead of regenerating it and hoping the good parts come back. When work moves past a single shot, the AI video editor takes it from there.
INPUT
Source clip
PROMPT
Remove the front basket from the bicycle.
OUTPUT VIDEO
0:04GETTING STARTED
How to Generate with MiniMax H3
Start from a prompt alone, or add images, clips, and audio for what you want locked. Working from a single frame? Use image to video instead.
Say what happens in order, and how each reference should be used. Note the sound you want — H3 fills in anything you leave out.
Choose an aspect ratio and a duration, then generate. The file comes back with its audio already mixed.
Ready to create your first video?
YouTube Videos on MiniMax H3 AI Video Model
I made a short film in ComfyUI — MiniMax H3 | THE RECITATION
CGlide (RedShade)
Short Movie - Minimax H3 2m20s (WF & Repo Soon)
CGlide (RedShade)
This 1-Minute AI Movie Looks TOO Real 🎬 (MiniMax H3, text-to-video)
Struggle Technology
MiniMax H3, the king of AI commercial ads
AnimAItion Studio
MiniMax H3's anime aesthetics are honestly on another level
AnimAItion Studio
MiniMax H3 on X: What People Are Making
Minimax h3 is just way too good. Rendered this locally on my machine. pic.twitter.com/PXkNL07zK4
— Jesse Wellens (@Jessewelle) August 20, 2026
MiniMax H3でテキストから15秒映像つくってみました。
— meropan (@renataro9) August 20, 2026
今回はちょっとノワール寄り。
色数を絞って、電車やバイク、街のシーンがテンポよくつながっています。
こういう世界観、H3かなり相性いいかも。
動きやカメラワークもいい感じ。
使用プロンプトは👇#MiniMaxH3 #MiniMaxDesig pic.twitter.com/lXariMenRM
【MINIMAX-H3】作り方とプロンプト📢
— SEIIIRU😈AIを使う映像クリエイター (@seiiiiiiiiiiru) August 14, 2026
最近AIが考えたプロンプトがよくシェアされていますが、長すぎて読む気すら失せる。
そんな方にオススメのやり方です👇(リプ欄) https://t.co/qwoZ25BFmJ pic.twitter.com/WKvmlFYquN
MINIMAX H3
— SEIIIRU😈AIを使う映像クリエイター (@seiiiiiiiiiiru) August 13, 2026
案件もろたので明日作り方公開しまーす。 pic.twitter.com/HgQBvvg4OZ
I think AI Agents are about to completely change how we prompt video models.
— Pierrick Chevallier | IA (@CharaspowerAI) August 19, 2026
I gave the MiniMax Design AI Agent one image and one simple instruction:
“Act as an expert FPV director and turn this into something viral.”
It analyzed the idea, built the camera logic, movement and… pic.twitter.com/t9eQP7hhZE
MiniMax H3
— Kōda (@aimikoda) August 13, 2026
Character Introduction Template
So it looks like MiniMax H3 understands parameters. You can rewrite the text parameters in the prompt to match your character and use it that way. I’m sharing the character sheet and the prompt I used below. pic.twitter.com/UgAoL4pF6c
Minimax H3 还是太强了
— 0xMomo (@0xmomonifty) August 21, 2026
简单尝试了一下,也能完美复刻最近流行的「大唐琵琶行」
步骤也非常简单:
> 原视频转成黑白深度视频
> 生成你想要替换的人物四视图
> 把原视频、深度视频和人物四视图发给 CodeX 让它帮你参考并生成对应的 H3 提示词
> 再把这几样多参以及提示词给到 Minimax 或者 ComfyUI… pic.twitter.com/Mvg83i3qcJ
🎬New MV【composition】
— tapehead.lab (@tapehead_Lab) August 6, 2026
新しいMVつくりました!
Music : Suno
Visual : Midjourney
Video Generation : MiniMax H3
MiniMax H3はオブジェクトをレイヤーとして処理する能力が高くて面白い!
プロンプトの通りも良すぎるので、いろんなタイプのプロンプト試すのが楽しいかもです🍿 pic.twitter.com/t6qlMKrUWi
AfterEffect → Minimax H3
— SEIIIRU😈AIを使う映像クリエイター (@seiiiiiiiiiiru) August 25, 2026
「AI動画をレンダリングマシンとして使う」
生成するとき…
・AEでラフな同コンテを作成
・Chat GPTでマテリアルを作成
・強化するプロンプトを作成
・H3でAEラフとマテリアル画像を参照
修正したいとき…
・AEで文字や動きの修正
・GPTでマテリアルを再生成… pic.twitter.com/HQLg8l0O0s
コピペで出来る!
— AIスタジオワンルーム(AIアニメ、動画、漫画) (@studio_oneroom) August 23, 2026
MiniMax H3
リファレンスしたキャラクターが指定した文字を操ります。
初期設定で入っている
文字列A = 「MiniMax H3」
文字列B = 「AI STUDIO ONEROOM」
だけ好きなものに変えればOKです。
-----------
プロンプト↓↓↓
-----------
# 【ここだけ変更】
文字列A = 「MiniMax… pic.twitter.com/CIj0nWumlV
コピペで出来る!
— AIスタジオワンルーム(AIアニメ、動画、漫画) (@studio_oneroom) August 25, 2026
モーショングラフィック!
MiniMax H3のプロンプトを作りました。
プロンプトの冒頭で好きな文字に変えて使えるようにしてあります。… pic.twitter.com/mNpCnZTuRZ
AI生成したキモい食べ物POPの動画作ってみた
— Hirokazu Yokohara (@Yokohara_h) August 24, 2026
画像はGPT、動画はローカルのMiniMax H3
個人的にある意味アートに見えてきてる pic.twitter.com/IkJuRpeJDn
FAQs
Make cinematic video with native stereo audio
Bring your references, write the brief, get a finished 2K clip with sound.