# MiniMax H3 Reference

MiniMax H3 Reference is the reference-to-video mode of MiniMax's H3 model, released on 31 July 2026. It builds a clip with native sound around your own images, videos and audio: a product, a person, a setting or a style. In Sloosh it takes up to 9 images, 3 videos and 3 audio clips and outputs up to 4K.

Web page: https://sloosh.ai/models/minimax-h3-reference

[Use MiniMax H3 Reference](https://sloosh.ai/studio/video?video-model=minimax-h3-reference)

[View templates](https://sloosh.ai/models/minimax-h3-reference/templates.md)

## Key facts in Sloosh

|   |   |
| --- | --- |
| Made by | MiniMax |
| Makes | video (with sound: always) |
| Inputs | text; up to 9 reference images; up to 3 reference videos (each 2 to 15 seconds, 15 seconds in total); up to 3 reference audio clips (only with another reference); 12 references in all; or a start frame (optional) instead of references; no end frame |
| Length | any whole number of seconds from 5 to 15 |
| Resolution | 480p, 768p, 2K (default), 4K |
| Aspect ratios | Auto, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 |
| Prompt limit | 50,000 characters |

## Best for

- Putting a real product into a new scene: MiniMax's omni-reference mode takes "up to 9 images, up to 3 video clips and up to 3 audio clips", up to 12 files in total ([MiniMax H3 model card](https://huggingface.co/MiniMaxAI/MiniMax-H3)).
- Combining a product, a person and a setting from separate files, each with its own role ([MiniMax reference guide](https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_ref_en.md)).
- Brand and e-commerce clips: MiniMax cites accurate text and brand rendering and V2V motion transfer ([MiniMax blog](https://www.minimax.io/blog/minimax-h3)).
- Matching a voice or soundtrack: audio references carry timbre, rhythm, music style or dialogue ([MiniMax reference guide](https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_ref_en.md)).

## How to use MiniMax H3 Reference in Sloosh

1. Open [Video](https://sloosh.ai/studio/video?video-model=minimax-h3-reference) and pick MiniMax H3 Reference in the model chip.
2. Add your references: images, videos and audio (audio needs at least one image or video with it). Or add a start frame instead; Studio closes the other option.
3. In the prompt, name each file by its number and give it a job ("Image 1 is the product").
4. Set length (5 to 15 seconds), resolution and aspect ratio (Auto follows your references). Set a seed to repeat a result.
5. Check the price on the Generate button, then generate.

## MiniMax H3 Reference prompting tips

- Number references within their own type: Image 1, Image 2, Video 1, Audio 1. MiniMax numbers each category separately ([MiniMax reference guide](https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_ref_en.md)).
- Give every file a job and say how much to keep, from fully kept to a loose reference. MiniMax's guide has four levels for visuals ([MiniMax reference guide](https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_ref_en.md)). A file with no stated job is the most common reason a result goes wrong ([Sloosh docs](https://sloosh.ai/docs/minimax-h3-reference.md)).
- Describe the relationship between the references and the target video in plain language ([MiniMax blog](https://www.minimax.io/blog/minimax-h3)).
- When you only want a voice's delivery, don't carry the reference's original words into the new clip ([MiniMax reference guide](https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_ref_en.md)).
- Keep sound, music and speech on separate lines, and stay under about 7,000 characters ([Sloosh docs](https://sloosh.ai/docs/minimax-h3-reference.md)).

## MiniMax H3 Reference example prompts

**Prompt 1**

```text
Image 1 is the product: a white ceramic table lamp with a linen shade. Keep its shape and colour exactly. Image 2 is the setting: a reading corner with a green armchair; use its light and layout only. One continuous shot: dusk falls and the lamp turns on, warming the room from Image 2. The camera arcs slowly to the right. Sound: a soft switch click, quiet room tone.
```

**Prompt 2**

```text
Image 1 is the presenter. Audio 1 is her voice; use its timbre and pace only. She stands in a bright studio and says, in English, cheerfully: "Three colours, one bag." The camera holds a medium shot. Music: light pop beat.
```

**Prompt 3**

```text
Video 1 is the motion to copy: a slow dolly along a shelf. Image 1 is the sneaker; keep its colours exactly. The camera follows the move from Video 1 and stops on the sneaker on the top shelf. Sound: store ambience.
```

## How it compares in Sloosh

|   | H3 Reference | H3 Max Reference | MiniMax H3 | Seedance 2.0 |
| --- | --- | --- | --- | --- |
| References (images / videos / audio) | 9 / 3 / 3 | 9 / 3 / 3 | none | 9 / 3 / 3 |
| Max resolution | 4K | 1080p | 4K | 4K |
| Length | 5 to 15 s | 5 to 15 s | 5 to 15 s | 4 to 15 s (fixed steps) |
| Sound | always | always | always | optional |

## Where else it runs

MiniMax's own API (Reference Generation mode of MiniMax-H3) and its Hailuo AI web app ([MiniMax video guide](https://platform.minimax.io/docs/guides/video-generation)).

## Frequently asked questions

### MiniMax H3 Reference vs MiniMax H3: when should I use references?

Use Reference to put a real product or person into a new scene. Use MiniMax H3 with start and end frames when the clip must open and close on exact images. One run can't use both.

### How many references does MiniMax H3 Reference take?

Up to 9 images, 3 videos and 3 audio clips in Sloosh, 12 in all.

### Why does it ignore or mix up my reference?

Usually the file has no stated job. Say what each file is for and how much to keep.

### Can MiniMax H3 Reference make 4K video in Sloosh?

Yes: 480p, 768p, 2K and 4K, starting at 2K.

### Does it make sound?

Yes, every clip has native sound.

### Can I use only an audio reference?

No. In Sloosh an audio clip needs an image or video reference with it.

## Related

- Models: [MiniMax H3](https://sloosh.ai/models/minimax-h3), [MiniMax H3 Max Reference](https://sloosh.ai/models/minimax-h3-max-reference), [Seedance 2.0](https://sloosh.ai/models/seedance-2.0), [Wan 3](https://sloosh.ai/models/wan-3.0)
- Docs: [MiniMax H3 Reference](https://sloosh.ai/docs/minimax-h3-reference.md)

## Sources

- [MiniMax H3 model card](https://huggingface.co/MiniMaxAI/MiniMax-H3)
- [MiniMax H3 reference prompt guide](https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_ref_en.md)
- [MiniMax: MiniMax H3](https://www.minimax.io/blog/minimax-h3)
- [MiniMax API: video generation guide](https://platform.minimax.io/docs/guides/video-generation)
- [Sloosh docs: MiniMax H3 Reference](https://sloosh.ai/docs/minimax-h3-reference)