# MiniMax H3 Reference MiniMax H3 Reference in Sloosh: H3 video with sound guided by up to 9 reference images, 5 to 15 seconds at up to 4K. Last updated: 2026-10-05 **Key facts** - MiniMax H3 Reference is the reference-to-video mode of MiniMax H3: it keeps a product, person, setting or style from up to 9 reference images. - MiniMax H3 Reference makes 5 to 15 second clips in Sloosh at 480p, 768p, 2K or 4K, with Auto and six aspect ratios. - MiniMax H3 Reference can't combine reference images with a start or end frame, because MiniMax keeps the two modes separate. - MiniMax H3 Reference always makes sound and has a seed setting in Sloosh. Web page: https://sloosh.ai/docs/minimax-h3-reference MiniMax H3 Reference makes H3 video from your references: a product, a person, a setting or a style. References and start or end frames can't be mixed; this option is the references one. ## When to use it - Putting a real product into a new scene and keeping it accurate. - Combining a product and a setting from separate images. ## How to prompt it - Name each file by type and order: "Image 1", "Image 2". - Give every file a job and say how much to keep: "keep its shape and label exactly", or "use its light and layout only". A file with no stated job is the most common reason a result goes wrong. - Write camera moves in plain English with size and speed: "The camera pushes in with small amplitude at slow speed." One move per continuous shot. - Say "one continuous shot, no cuts" for a single take; otherwise "then" can turn into a cut. - Keep sound, music and speech apart: "Sound: ...", "Music: ...". Put spoken lines in quotes with the speaker, language and tone. Every clip has sound. - Write constraints as plain sentences; there's no negative prompt. Stay under about 7,000 characters. ## Example Two references with stated jobs: the product to keep exactly, and the setting to take light and layout from. Input: product photo: ![An amber glass serum bottle with a black dropper and a cream label reading SERUM 01, on a light grey background](https://sloosh.ai/docs/in-bottle.webp) -> Input: setting: ![An empty white marble bathroom counter by a window with sheer curtains and a small fern](https://sloosh.ai/docs/in-counter.webp) -> Result: [![The serum bottle on the marble counter in moving morning light, the camera arcing to the label](https://sloosh.ai/docs/m-minimax-h3-reference.webp)](https://sloosh.ai/docs/m-minimax-h3-reference.mp4) (video) _16:9, 768p, 6 seconds, Enhance off. Clips here play without sound; press one to pause._ **Prompt** ```text Image 1 is the product: an amber glass serum bottle with a black dropper and a cream "SERUM 01" label. Keep its shape and label exactly. Image 2 is the setting: a white marble bathroom counter by a window. Use its light and layout only. One continuous shot: the bottle stands on the counter from Image 2 as morning sunlight moves slowly across it. The camera arcs to the left with small amplitude at slow speed, ending on a close-up of the label. Sound: soft room tone and distant birdsong. Music: a slow, sparse piano melody. ``` ## Settings in Sloosh - Aspect ratio: Auto, 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16. - Resolution: 480p, 768p, 2K or 4K. - Length: 5 to 15 seconds. - References: up to 9 images. - Start frame: optional; no end frame. - A start or end frame can't be combined with reference images. - Sound: always on. - Seed: available, to repeat a result. - Prompt: up to 50,000 characters. Open it in [Video](https://sloosh.ai/studio/video) and pick MiniMax H3 Reference in the model chip. The Generate button shows the cost first. ## Frequently asked questions ### MiniMax H3 Reference vs MiniMax H3: when should I use references instead of frames? Use MiniMax H3 Reference to put a real product or person into a new scene, where the model is free to choose the framing. Use [MiniMax H3](https://sloosh.ai/docs/minimax-h3.md) with a start and end frame when you need the clip to open and close on exact images. MiniMax doesn't allow both in one run. ### How many reference images does MiniMax H3 Reference take? MiniMax H3 Reference takes up to 9 reference images in Sloosh. Name each one by order in the prompt, "Image 1", "Image 2", and give every image a job, such as the product to keep exactly or the setting to take light from. ### Why does MiniMax H3 Reference ignore or mix up my reference? The most common cause is a reference with no stated job. Say what each image is for and how much to keep: "keep its shape and label exactly", "loosely", "only the colour" or "only the mood". Those four levels match MiniMax's own reference guide. ### Can MiniMax H3 Reference make 4K video in Sloosh? Yes. MiniMax H3 Reference offers 480p, 768p, 2K and 4K in Sloosh, and starts at 2K. 4K outputs need a Pro or Max plan; [Plans and credits](https://sloosh.ai/docs/plans-and-credits.md) explains plans. For faster clips at up to 1080p, use [MiniMax H3 Max Reference](https://sloosh.ai/docs/minimax-h3-max-reference.md). ## Sources - [MiniMax H3 prompt guide](https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_base_en.md) - [MiniMax H3 reference guide](https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_ref_en.md) - [MiniMax: H3 announcement](https://www.minimax.io/blog/minimax-h3)