Priya Kapoor
Senior Creative Producer
"We used to shoot a rough cut, then spend a day just getting the resolution up to client standard. MiniMax H3 outputs 2K straight out of generation, so that whole step disappeared from our workflow."
Models
MiniMax H3 40/80/120
2K video generation from text, images, or reference media with flexible 4–15s duration.
Omni Reference(Optional)
First/End Frame
Drag or upload Media (0/15)
Image, Video, Audio (max 64MB)
Prompt
Use @ to mention a reference. Example: Make the character in @Image 1 dance using the moves in @Video 1 and add the music from @Audio 1.
0/5000
Auto
4s
480P
1
Credits required: 4
MiniMax H3 (Hailuo 3.0)
Also known as Hailuo 3.0. Generate native 2K video with synced audio from text, image, video, and audio references — then edit any clip by instruction, no regeneration needed.
MiniMax H3 (Hailuo 3.0)
MiniMax H3, also known as Hailuo 3.0, creates native 2K AI video with synced audio. Combine image, video, and audio references, edit any clip by instruction.
Try MiniMax H3 Free
What is MiniMax H3?
MiniMax H3, also known as Hailuo 3.0, is MiniMax's newest general-purpose multimodal video model. It reads text, images, video clips, and audio together in a single generation, and outputs native 2K video in one pass. Unlike models that take one input at a time, H3 accepts up to 9 reference images, 3 reference videos, and 3 reference audio clips at once — up to 12 files combined — so a character's look, a camera style, and a voice can all come from your own references instead of being described in words alone. And when a finished clip needs one change, H3 supports instruction-based editing: describe the adjustment, and the model updates that part directly instead of starting the whole generation over.
Why Use MiniMax H3?
Whether you're producing a batch of UGC-style product ads or a multi-shot brand trailer, film-opening title sequence, or vertical drama scene, MiniMax H3 is built to hold one consistent look across every cut. Feed it a product photo, a brand color reference, and a voice sample, and H3 keeps that identity steady from the first frame to the last — at native 2K, with dialogue, sound effects, and ambience generated in the same pass. For teams that need to move between formats in the same week — an ecommerce ad on Monday, a trailer cut on Wednesday, a drama scene on Friday — one model that keeps the same face, the same voice, and the same visual style across every version changes how much a single production day can cover.
When 1080p Isn't Enough — and Neither Is Starting Over
Your video looks almost right: the product is crisp, the delivery is close, the composition works — except the resolution caps out below what a client-facing cut needs, and the audio still needs a separate pass before it sounds finished. Then one detail is off — a prop's color, a line of dialogue, a beat in the pacing — and most models give you exactly one option: regenerate the whole clip and hope everything else still holds up. MiniMax H3 is built against both problems at once. It outputs native 2K with dialogue, sound effects, and ambience generated together in the same pass, so there's no separate audio session and no downgraded resolution to explain to a client. And because it supports instruction-based editing, a single wrong detail is a text instruction away from fixed — not a full re-roll of the whole generation.
What Makes MiniMax H3 Stand Out
From a single product photo to a multi-shot brand trailer — MiniMax H3 is built for teams that need native 2K quality, consistent references across every shot, and the ability to fix a detail without regenerating from scratch.
Native 2K Video Generation
MiniMax H3 outputs 2K resolution in every generation — a step up from the 1080p ceiling of earlier models. Clips run 5 to 15 seconds, giving you footage sharp enough for a hero product shot or a trailer close-up without an extra upscaling pass.
Omni-Reference Control Across Every Shot
Combine up to 9 reference images, 3 reference videos, and 3 reference audio clips — up to 12 files in one generation — to hold a character, a camera style, or a voice consistent across a full production. Reference a face, an outfit, a walk, and a voice sample together, and MiniMax H3 reads them as one brief instead of forcing a single input to carry everything.
Instruction-Based Editing, No Regeneration
When a generated clip is close but one detail is off, describe the change instead of starting over. Adjust a character, an object, a background, or the pacing directly on the existing clip, and MiniMax H3 applies the edit while keeping the framing and performance you already approved.
How to Use MiniMax H3
No camera, no crew, no separate audio session. Upload your references, write your brief, and MiniMax H3 delivers native 2K video with synced audio — ready to review, or refine further by instruction.
Step 1: Upload Your References
Add a first-frame or last-frame image to anchor the shot, or combine up to 9 reference images, 3 reference videos, and 3 reference audio clips to guide character, style, camera, and voice. Reference audio must be paired with an image or video — it can't be submitted on its own.
Step 2: Write Your Prompt and Set Duration
Describe the scene, the action, and the camera movement in your prompt (up to 7,000 characters). Choose a duration between 5 and 15 seconds — MiniMax H3 outputs native 2K on every generation.
Step 3: Generate, Review, and Edit by Instruction
MiniMax H3 generates your video with synchronized dialogue, sound effects, and ambience in one pass. If a detail needs adjusting, describe the change directly — no need to regenerate the whole clip from scratch.
Create Your First MiniMax H3 Video Today
From a single product photo to a multi-shot brand trailer or short-form drama scene, MiniMax H3 delivers native 2K video with synced audio in one generation — and lets you fix any detail by instruction instead of starting over. Built for UGC ad teams scaling output, and for brand, trailer, and drama teams that need the same character, voice, and style held steady across every cut.
Try MiniMax H3 FreeWhat Users Are Saying About MiniMax H3
Daniel Osei
Post-Production Lead
"The instruction-based editing changed how we revise. One note from a client used to mean a full re-generation and a coin flip on whether the rest of the clip still matched. Now we just describe the fix."
Nina Fontaine
Brand Video Director
"Feeding in our brand references — the product, the model, the voice sample — together in one generation is the difference between 'looks kind of like our brand' and actually matching it."
Marco Delgado
Short-Form Drama Showrunner
"Our vertical drama scenes need the same two characters to look right across a dozen cuts. The omni-reference setup is the first workflow where that's held up consistently across a full episode."
Yuki Tanaka
Performance Marketing Lead
"Native audio at 2K meant we stopped sending clips out for a separate sound pass before testing. That alone cut a day off our creative turnaround."
Chloe Bennett
Trailer Editor
"For a teaser cut, resolution is the first thing anyone notices. MiniMax H3 is the first model where we didn't have to explain why the trailer looked softer than the feature it was cut from."
MiniMax H3 — Frequently Asked Questions
MiniMax H3 is MiniMax's newest multimodal video generation model. It generates native 2K video from text, images, video clips, and audio references, with synchronized dialogue, sound effects, and ambience in a single pass, plus instruction-based editing for revising a finished clip.
Yes. MiniMax H3 is the model's official name; Hailuo 3.0 is how it's commonly referred to as the next generation of MiniMax's Hailuo video line.
MiniMax H3 outputs native 2K resolution. Each generation runs 5 to 15 seconds, in integer values, with common aspect ratios or an adaptive ratio based on your input.
Three modes are supported: Text-to-Video (a prompt alone), First/Last-Frame Image-to-Video (a prompt plus a starting and/or ending frame), and Reference Generation (a prompt plus reference images, video, or audio to guide character, motion, camera, style, or voice).
Yes. MiniMax H3 accepts up to 9 reference images, 3 reference video clips, and 3 reference audio clips in a single generation, with a combined limit of 12 reference files. Reference audio must be submitted together with an image or video — it can't be sent on its own.
Yes. Dialogue, sound effects, and ambient atmosphere are generated in the same pass as the video, timed to the on-screen action.
Yes. MiniMax H3 supports instruction-based editing. Describe the change you want — to a character, an object, a background, or the pacing — and the model updates that part of the clip directly, instead of requiring a full regeneration.
Seedance 2.0 is built around multi-shot UGC ad structure — a hook, a product moment, and a call to action in one generation. MiniMax H3 is built for native 2K output, broader reference control (up to 9 images, 3 videos, and 3 audio clips at once), and instruction-based editing on a finished clip — useful whether you're producing UGC ads, brand trailers, or short-form drama.
Video references support H.264/AVC and H.265/HEVC (in-video audio as AAC or MP3); images support JPG, JPEG, PNG, WEBP, HEIC, and HEIF; audio supports WAV and MP3. Size limits are up to 50MB per video file, 30MB per image, and 15MB per audio file.
Prompts can run up to 7,000 characters.
Reference Generation lets you submit a prompt together with reference images, videos, or audio — in any combination — and MiniMax H3 keeps the features of that reference subject or asset consistent throughout the generated video.
MiniMax H3 fits UGC ad teams that need higher-resolution output at scale, and it extends naturally to brand video teams, trailer editors, and short-form drama producers who need a consistent character, voice, and visual style held across every cut.