MiniMax H3 Max Reference to Video from Images, Clips & Audio

Last verified: October 29, 2026

MiniMax H3 Max can take a character, a visual style, a motion, a camera move, a voice or an editing rhythm from files you supply and build a new shot around them. With MiniMax H3 Max Reference to Video, you put stills in Add Image, add a clip or a sound file if the shot needs one, spell out in the Prompt what each file should contribute, and set Aspect Ratio to 21:9 when the scene calls for an ultrawide, letterboxed frame.

Writing a scene from scratch can't give you this much control over what stays fixed. H3 Max reads the text prompt together with reference images, videos and audio in whatever mix the shot needs. A face can come from one photo, the look from another and the movement from a clip, while the prompt describes the action that ties them together.

That makes the generator a good fit for recurring characters, mascots and brand looks that have to match across a campaign, and for shots you want to stage like existing footage. Images anchor every run. Clips and sound are optional layers you add when the shot needs movement or a voice taken from real material.

Explore more reference to video

Reference To Video Use in Text To Video Use in Image To Video Use in First To Last Frame Seedance 2.5 MiniMax H3 Wan 3.0 Prime Kling 4.0 Vidu Q4 Gemini Omni 1.1 Flash
Capability Snapshot

Inputs, Clip Lengths and Formats at a Glance

What this generator accepts and what it returns, in one view.

Input

Up to 9 images (JPG, JPEG, PNG or WEBP, up to 10 MB each) + up to 3 videos (MP4 or MOV, up to 50 MB each) (optional) + up to 3 audio files (MP3 or WAV, up to 15 MB each) (optional) + a text prompt of up to 7,000 characters

Video Duration

5 to 15 seconds

Resolution

480P, 768P or 1080P

Aspect Ratio

21:9, 16:9, 4:3, 1:1, 3:4, 9:16

Output

1 video per run

Techniques for Stronger Reference Results

Assign One Role to Every Upload

H3 Max can draw a character, a style, a motion, a camera move, a voice or an editing rhythm from a reference, so a file with no stated job leaves an open question in your brief. Open the Prompt with one short line per upload. Identify each file by what it shows, for example 'the photo of the woman in the green coat supplies her face and clothing; the clip supplies only the slow push-in, not its location'. Be as clear about what to ignore as about what to keep.

Keep the Look Separate from the Subject

Style is a reference use of its own, so give the look its own image instead of hoping it carries over from the character photo. Upload a frame that shows the grade, texture or art style you want next to your subject images, and say in the Prompt that it controls only the look. You can then swap the style frame between runs while the character references stay the same.

Borrow the Cut Pacing of an Existing Edit

Editing rhythm is another thing H3 Max can take from a reference. That helps with trailers, product spots and music-led pieces, where the timing of the cuts matters as much as the images. Add a clip whose pacing you like in Add Video, describe your own shots in the Prompt in the order they should play, and state that the clip sets only the rhythm. Then set Video Duration close to the clip's length so the pacing has the same room.

How to Make a Video from Your References with H3 Max

Six steps from your reference files to a downloaded clip.

Choosing Between H3 Max and MiniMax H3 for Reference Work

When the fast variant is enough and when MiniMax's standard H3 model is the better choice.

Turnaround while testing references H3 Max is MiniMax's fast generation variant and generates faster than MiniMax H3, which suits several quick rounds of reference tests. MiniMax H3 is the variant MiniMax positions for high-quality video creation, and it also offers reference-to-video. Pick H3 Max when you expect many passes to get each reference doing its job, and MiniMax H3 when one high-quality final render matters more than speed.
Top output resolution On Vidofy, H3 Max's highest Resolution setting is below 2K, which covers web and social delivery. MiniMax H3 renders at 768P or 2K. Stay with H3 Max for clips headed to web and social feeds, and move to MiniMax H3 when the delivery spec calls for a 2K master.
Use MiniMax H3 Max when you are iterating on reference setups and delivering for web or social; switch to MiniMax H3 when a 2K master is the priority.

Common Problems with Reference Uploads and Clips

Upload, length and prompt problems that come up in reference-to-video runs, and what to change.

My reference clip won't upload or gets rejected.

Add Video only accepts its listed formats under a size cap, and it applies two time limits: one per clip and one for all clips together. Three clips that each pass can still be refused as a set. Trim each clip down to the part with the motion or camera move you want, re-export it in an accepted format if needed, and upload again. The exact limits are in the FAQ below.

My image is rejected even though it looks fine.

Add Image checks the format, the file size and a minimum pixel size. A small thumbnail, a large uncompressed export or a phone photo saved in a format outside the accepted list will be refused. Re-save the image at full resolution in one of the accepted formats, compress it if it is over the size cap, and add it again.

My voice recording won't load in Add Audio.

Phone voice memos are often saved in formats Add Audio does not accept, and long recordings run past its per-file and combined time limits. Convert the file to an accepted audio format, cut it down to the stretch with the clearest speech, and upload it again. A short, clean sample is also easier to point to in the Prompt.

I need a clip shorter or longer than Video Duration allows.

H3 Max does not render four-second clips, and each run here returns a single clip within the Video Duration range. For a shorter beat, generate at the shortest setting and trim it in your editor. For a longer piece, plan several shots that reuse the same reference set and join them afterwards.

The Prompt box stops accepting text.

The Prompt box has a character limit, and a detailed brief with several references can reach it. Keep the lines that give each upload its role, cut repeated adjectives and anything your references already show, then paste the shorter version back in and generate.

Frequently Asked Questions

Which files can I upload, and what are the size and length limits?

Add Image is required and takes 1 to 9 images in JPG, JPEG, PNG or WEBP, each up to 10 MB and at least 300×300 pixels. Add Video is optional and takes up to 3 MP4 or MOV clips of up to 50 MB each; each clip can be up to 15 seconds long, and all clips together can't exceed 15 seconds. Add Audio is optional too: up to 3 MP3 or WAV files of up to 15 MB each, with the same 15-second limit per file and in total.

Can I use just a video or just an audio file as the reference?

No. Every run needs at least one image in Add Image, and clips and audio files are optional additions on top of it. If all you have is footage of a character, export a clear still from it, upload that as the image, and add the footage as the clip.

What makes one generation cost more than another?

The credit cost is calculated from the options you choose and shown above the Generate button once you're signed in. It changes with Video Duration, Resolution, the length of any videos you upload and the number of images you add. The Enhance Prompt button doesn't use credits.

Why doesn't my video open on my reference image?

Reference images aren't meant to be an exact starting frame. In MiniMax's documentation, setting an exact first or last frame belongs to the model's image-to-video mode, and reference input is listed as a separate way of working. Treat your images as the source of the subject and the look, and describe the opening composition in the Prompt when it matters.

References

Sources and citations used to support the content provided above.

Updated: 2026-10-29 21:15:51 3 Sources

platform.minimax.io

Source Link
https://platform.minimax.io/docs/guides/video-generation

platform.minimax.io

Source Link
https://platform.minimax.io/docs/api-reference/video-generation-v2-create

platform.minimax.io

Source Link
https://platform.minimax.io/docs/api-reference/api-overview