MiniMax H3 Text-to-Video
Describe the scene, shots, dialogue and sound in up to 7,000 characters. Pick 4–15 seconds, 768p or 2K, and 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16.
Open workspaceMiniMax H3 AI Video Generator: 2K Videos with Native Stereo Sound
MiniMax H3, also known as Hailuo 03, is MiniMax’s general-purpose multimodal video model. It understands text, images, video and audio in one creative context and makes 4–15 second videos at up to 2K with native stereo sound. MiniMax H3 follows detailed directions for characters, action, camera, shots, style, dialogue, music, readable text and brand elements.
Text, frames or references to video · 768p or 2K · 4–15 seconds · stereo sound
MiniMax H3 Text, Frames and Reference Workflows
Three workflows share 4–15 second lengths, 768p or 2K quality and stereo sound. Start from words, from keyframes or from images, video and audio.
MiniMax H3 First and Last Frame
Upload a JPG, PNG or WEBP first frame up to 30 MB and, if you like, a last frame. Each side must be 256–5760 px and the shape 0.4–2.5.
Open workspaceMiniMax H3 Reference-to-Video
Add up to 9 reference images, an optional MP4 or MOV clip of 2–15 seconds and up to 3 MP3 or WAV clips of 15 seconds combined, then call them Image1, Video1 and Audio1 in the prompt.
Open workspaceExplore MiniMax H3 Features
MiniMax H3 brings multimodal context, native 2K and stereo sound, multi-reference control, precise instructions, keyframes and text-to-video scenes to one model.
Unified Multimodal Context in MiniMax H3
MiniMax H3 reads text, images and audio within the same context. Each input can play its own role, such as setting a character's identity, a product's look, a visual style or the voice that sings. Just describe in words how each file relates to the video, for example “the character in Image1 sings the song from Audio1”. MiniMax H3 keeps all of these creative elements connected in the final clip.
- Text, images and audio in one context
- Give every file its own role
- Describe the links in plain words
Native 2K Video and Stereo Sound with MiniMax H3
MiniMax H3 generates up to 2K video together with native stereo sound, instead of adding audio in a separate step. Action, dialogue, music, ambience and sound effects can all be directed in the same prompt. That gives stronger audio-visual continuity and more natural timing between what you see and what you hear. MiniMax offers 2K by default, and MiniMax H3 also renders at 768p for quicker, lower-cost drafts.
- Up to 2K, or 768p for drafts
- 32 kHz stereo sound made with the picture
- Dialogue, music and effects in one prompt
Flexible Multi-Reference Control in MiniMax H3
Combine several images, a video clip and audio in one generation. Identity can come from one image, a product or outfit from another, motion from the video and a voice or sound from the audio reference. Natural-language instructions explain how every file should shape the result, which reduces conflicts between references. On this site MiniMax H3 accepts up to 9 reference images, one 2–15 second video clip and up to 3 audio clips of 15 seconds combined.
- Up to 9 reference images
- One video clip and up to 3 audio clips
- Name them Image1, Video1 and Audio1
Strong Instruction Following and Text Rendering in MiniMax H3
MiniMax H3 follows detailed directions for shot order, timing, transitions, camera movement, character behavior and sound design. It also renders clearer text, titles, interface elements, signs and brand details. MiniMax H3 responds reliably both to what you ask for and to what you rule out, such as elements that must stay unchanged. That makes it a good fit for opening titles, animated posters and product films.
- Shot order, timing and camera on cue
- Clearer titles, signs and brand details
- Respects what must stay unchanged
Animate an Image or First and Last Frames with MiniMax H3
Image to video works with one uploaded image or with a pair of first and last frames. A single image sets the opening for natural character motion, moving scenery and camera animation. Adding a closing frame gives you clearer control over how the clip begins, changes and ends. MiniMax H3 works well for animated posters, product showcases, interface demos, character moments and key-art animation.
- One image or a first and last frame
- Shape follows your image
- JPG, PNG or WEBP, 256–5760 px per side
Build Original Audio-Visual Scenes from Text with MiniMax H3
Text to video creates a complete clip with sound straight from a written prompt, with no uploads needed. Describe the setting, characters, action, camera, shot progression, look, dialogue, music and ambient sound in one go. MiniMax lists stable dialogue in 11 languages, including English, Chinese, Japanese, Korean, Spanish, French, German, Portuguese and Russian. Use MiniMax H3 for ad concepts, cinematic sequences, animated shorts, motion graphics and social content.
- No uploads needed
- Stable dialogue in 11 languages
- 21:9 to 9:16 aspect ratios
How to Create Videos with MiniMax H3
Make a complete clip with sound in MiniMax H3 in three simple steps:
MiniMax H3: Choose a workflow and add inputs
Sign in and open the MiniMax H3 workspace. Write a prompt for text-to-video, upload a first frame and an optional last frame, or add up to 9 reference images, one 2–15 second video clip and up to 3 audio clips.
MiniMax H3: Describe the scene and pick settings
Write the action, shots, timing, dialogue, music and sound, give every reference a clear role, and say what must stay unchanged. Choose 4 to 15 seconds, 768p or 2K, and an aspect ratio. Check the credit estimate before you start.
MiniMax H3: Generate, review and download
Start the generation and MiniMax H3 creates your clip with picture and stereo sound together. Preview it in your library, refine the prompt, frames or references if needed, and download the finished video for ads, product pages or social posts.
Model comparison
MiniMax H3 vs MiniMax Hailuo 2.3: What's New
MiniMax lists Hailuo 2.3 as a legacy text and image to video model. MiniMax H3 adds native stereo sound, 2K output, lengths from 4 to 15 seconds and multimodal references in one general-purpose model.
| Capability | MiniMax H3Up to 2K | MiniMax Hailuo 2.3 |
|---|---|---|
| Positioning | General-purpose multimodal video model | Text and image to video model |
| Workflows | Text, first and last frame, multimodal reference | Text to video and image to video |
| Maximum quality | 2K | 1080p |
| Quality options | 768p or 2K | 768p or 1080p |
| Video length | Any length from 4 to 15 seconds | 6 or 10 s at 768p, 6 s at 1080p |
| Native sound | 32 kHz stereo, always on | Not listed |
| Frame rate | 24 fps | 24 fps |
| First and last frame | Yes | Not listed |
| Reference inputs | Text, images, video and audio | Not listed |
| Dialogue languages | 11 languages with stable support | Not listed |
| Aspect ratios | 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 | Not listed |
| Known for | Instruction following, text and brand rendering | Instruction following and physics |
MiniMax H3 Frequently Asked Questions
What is MiniMax H3?
MiniMax H3 is a general-purpose multimodal video model from MiniMax. It understands text, images, video and audio in one context and generates 4–15 second videos at up to 2K with native stereo sound.
Is MiniMax H3 the same as Hailuo 03?
Yes. MiniMax H3 is also known as Hailuo 03 or Hailuo 3.0. It follows Hailuo 01 and Hailuo 02 and joins text, image, video and audio tasks in one model instead of separate expert models.
Which workflows does MiniMax H3 offer here?
In MiniMax H3 you can start from a text prompt, animate a first frame with an optional last frame, or combine up to 9 reference images and a video clip with optional audio. All three make videos with stereo sound.
How long can MiniMax H3 videos be, and in what quality?
Choose any length from 4 to 15 seconds; the default is 6. MiniMax H3 renders at 768p or 2K at 24 fps, with 2K selected by default here. Text-to-video and references offer 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16, and frames follow your image.
Does MiniMax H3 generate sound?
Yes. Every MiniMax H3 video comes with a 32 kHz stereo track made together with the picture, and sound cannot be turned off. Describe dialogue, music and sound effects in the prompt; MiniMax lists stable dialogue in 11 languages.
How many credits does MiniMax H3 use?
MiniMax H3 is billed per second of output: 16 credits at 768p and 26 credits at 2K. A 6-second 2K video uses 156 credits, a 15-second 2K video uses 390, and a 6-second 768p draft uses 96. A reference video adds its own seconds at the same rate, so a 5-second clip with a 6-second 2K video uses 286. The first 5 reference images are included and each further image adds 8 credits; audio adds nothing.
How many reference images, videos and audio clips can I use?
MiniMax H3 here takes up to 9 reference images, one video clip and up to 3 audio clips. Add at least one image or the video; audio cannot be used alone. Call the files Image1, Video1 and Audio1 in the prompt and say what each one controls.
Which image, video and audio files can I upload?
For MiniMax H3, images can be JPG, PNG or WEBP up to 30 MB, with each side from 256 to 5760 px and an aspect ratio from 0.4 to 2.5. A reference video can be MP4 or MOV up to 50 MB and 2–15 seconds, with each side from 256 to 5760 px and 24 to 60 fps. Audio can be up to 3 MP3 or WAV files up to 15 MB each, 2–15 seconds each and 15 seconds combined. The upload box tells you if a file does not fit.
Can I use reference videos or edit existing footage?
Yes. In MiniMax H3 you can add one reference video of 2–15 seconds to guide motion, camera or a scene, and its seconds are billed with the output. Editing existing footage and using more than one reference video are not offered on this site yet.
How is MiniMax H3 different from Hailuo 2.3?
MiniMax lists Hailuo 2.3 as a legacy model for text and image to video at 768p or 1080p and 6 or 10 seconds. MiniMax H3 adds native stereo sound, 2K output, any length from 4 to 15 seconds and multimodal references.
How do I write better prompts for MiniMax H3?
Organize the prompt around visual intent, reference roles, timing, camera, sound and what must stay the same. Give every reference one clear job, describe shots in order, and add positive instructions plus explicit limits so MiniMax H3 follows one creative plan.