MiniMax
MiniMax H3 Reference: AI video generation
MiniMax H3 video guided by reference images, videos and audio, at 768p or 2K, from 4 to 15 seconds.
Video is paid per clip from your wallet, no subscription. The studio shows the price before you pay, and top ups start at $5.

Key features
MiniMax H3 Reference turns reference images into video inside Enhance AI. It produces clips from 4 to 15 seconds at a resolution you choose, 768p by default. On top of that it renders 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16 and it accepts up to 4 reference images. MiniMax H3 Reference is built by MiniMax. On Enhance AI it runs on the same account as every other model, with no separate subscription.
Clip length
MiniMax H3 Reference renders clips from 4 to 15 seconds, and you pick the length before you render. Longer clips cost more because video is billed by length, so rough out an idea short and only render the full length once the shot works.
Resolution
Choose 768p or 2K before you render. A lower resolution costs less, so test a prompt low and rerender the keeper at the top setting. The studio shows the price for the setting you picked before anything is charged.
Input modes
MiniMax H3 Reference takes a written prompt, reference images, reference videos (optional) and reference audio (optional). Add them in the studio, check the preview and render. Every output lands in your history, ready to download or to pass to another model on the same account.
Reference images
Add up to 4 reference images and MiniMax H3 Reference holds a character, a product or a style steady across shots: the gap between a clip that looks like your brand and one that looks like the model's idea of it. Attach them next to the prompt.
Aspect ratios
MiniMax H3 Reference renders 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16. Pick the ratio for the platform before you render rather than cropping afterwards: 9:16 for Reels and Shorts, 16:9 for YouTube, 1:1 for feeds and 21:9 for cinema. The whole frame is generated for that shape.
Specifications
- Provider
- MiniMax
- Type
- Video
- Released
- 31 July 2026
- Resolution
- 768p by default
- Duration
- 4 to 15 seconds
- Aspect ratio
- 21:9, 16:9, 4:3, 1:1, 3:4, 9:16
- Reference images
- Up to 4
How to use MiniMax H3 Reference on Enhance AI
- 1
Sign in and pick MiniMax H3 Reference
Open the Video Studio on Enhance AI and choose MiniMax H3 Reference from the model list. Video is paid per clip from your wallet, which starts empty on a new account, so top it up before your first render. The Try button on this page opens the studio with MiniMax H3 Reference already selected.
- 2
Add references and describe the shot
Upload your reference images (up to 4) and describe the shot they should appear in: the setting, the action and the camera move. Then pick a length from 4 to 15 seconds, a resolution and an aspect ratio.
- 3
Generate, then download or extend
Press Generate. MiniMax H3 Reference renders in the cloud and the clip lands in your history on Enhance AI, where you can download the MP4, upscale it or extend it with another model on the same account. You see the price before you confirm.
What people make with MiniMax H3 Reference
A product across scenes
Upload photos of the product and describe each setting, and MiniMax H3 Reference places the same bottle, shoe or device in every shot with its shape and label intact. Build a full campaign of scenes from a single set of studio photos rather than shooting each location.
Style from a mood board
References do not have to be a subject. Feed MiniMax H3 Reference frames from a film, a set of photographs or your own past work, and the model carries the palette, the grain and the framing into new shots. It is the fastest way to match a brand look or a client's reference reel.
Matching an existing shot
When a new clip has to cut against footage you already have, give MiniMax H3 Reference a frame from that footage as a reference. The new shot picks up the light, the lens and the colour of the original, so the join in the edit is invisible rather than a visible change of source.
Common questions
- What is MiniMax H3 Reference?
- MiniMax H3 Reference is a reference to video model from MiniMax. It builds a new shot from up to 4 reference images and a written prompt. On Enhance AI it runs on the same account as 250+ other models, with no separate subscription.
- What do I need to upload to use MiniMax H3 Reference?
- A written prompt, reference images, reference videos (optional) and reference audio (optional). Add them in the studio on Enhance AI, check the preview and render. The output lands in your history, ready to download or to pass to another model on the same account.
- When was MiniMax H3 Reference released?
- MiniMax released MiniMax H3 Reference on 31 July 2026. You can run it on Enhance AI now, along with every other video model on one account.
- How long can a MiniMax H3 Reference clip be?
- MiniMax H3 Reference produces clips from 4 to 15 seconds. Longer runs cost more, because video is billed by length rather than by generation.
- What resolution does MiniMax H3 Reference output?
- MiniMax H3 Reference renders at 768p or 2K. You pick the resolution before you render, and a lower one costs less.
- Does MiniMax H3 Reference take reference images?
- Yes, up to 4. Use them to hold a character, a product or a style steady across shots.
- Is MiniMax H3 Reference free to try?
- No. Video on Enhance AI is paid per clip from your wallet, and a new account starts with an empty wallet, so there are no free MiniMax H3 Reference clips. The free tokens you get on signup, with no card needed, pay for image and audio tools, not video. Wallet top ups start at $5.
- What does it cost to run MiniMax H3 Reference?
- MiniMax H3 Reference renders are billed from your wallet by clip length, resolution and the length of the file you upload, so a short clip at a lower resolution costs less than a long one at full quality. You see the price for your settings before you render.
- Do I need a separate subscription for MiniMax H3 Reference?
- No. MiniMax H3 Reference runs on the same Enhance AI account as every other model, so you are not paying one provider for video and another for images.
More from MiniMax
- MiniMax H3MiniMax H3 is a 2K video model that turns a prompt or a first frame into coherent, high motion video, with durations from 5 to 15 seconds.View
- MiniMax H3 MaxMiniMax H3 Max turns a prompt or a first frame into video at up to 768p, from 5 to 15 seconds, with a prompt expansion mode that fills in detail the prompt leaves out.View
- SpandaAnimate a photo into video with native stereo audio. Up to 15 seconds, 480p to 1080p.View
- H3 Video EditRewrite the lighting, style, weather or elements of a video from a prompt, with native audio. 480p or 768p.View
- H3 Video ExtendContinue a video from its last frame with a new segment and native audio. 480p or 768p, 3 to 15 seconds.View
- Spanda ReferenceBuild a video from up to nine reference images, three videos and three audio clips, with native stereo audio.View
Other reference to video models
More from Enhance AI