Log inUpgrade

MiniMax H3 prompts that actually ran

Every prompt below was generated with MiniMax H3 in Made, and the output next to it is what came back — not a curated stock image. Copy a prompt as a starting point, then use the guide underneath to adapt it.

Original results

MiniMax H3 prompts and their outputs

These examples use media produced for Made. Prompts are included when the production record allows them.

Action and sound

“Vertical phone video from the back seat of a small car rolling through an automatic tunnel car wash, 0.5x ultra-wide, handheld. Two friends in their mid-20s sit up front. The driver, a guy in a bright orange puffer jacket, ducks as a brush hits his window. The passenger, a Korean woman with a high ponytail in a lilac zip hoodie, twists around toward the camera screaming and laughing, her face bright and clear, no freckles. Huge spinning blue brushes slam across the windshield and side windows, thick pink foam slides down the glass in sheets, water jets drum on the roof, and the car rocks on the conveyor. Loud whirring brushes, hammering water, screams and laughter. White tunnel lights flicker through the foam, slight motion blur, muted true colors with the pink foam as the one accent.”

Image-to-video that talks

“She looks straight into the camera and says, "It wasn't even dirty. I just like the pink foam." Then she scoops a handful of foam off the hood and blows it at the lens, laughing. Foam slides down the windshield behind her, a pressure washer hisses in the next bay, the fluorescent tubes buzz. Handheld phone camera with a slight wobble, direct flash.”

8 second scene with a line

“Handheld phone video on a bright, harsh afternoon at the exit of an automatic tunnel car wash, 0.5x wide. The air dryers roar inside the empty tunnel, and out through the steam walks a sheep the dryers have puffed up to three times its normal size: a perfectly round ball of fluffy white wool as tall as a man's chest, with its small black face and thin black legs poking out. It trots onto the wet concrete, stops and gives a big shake, the wool bouncing like a cloud. The camera swings right to the attendant, a Latino guy in his late 20s in a plain bright yellow rain jacket and rubber boots, his face clear and bright in the sun, no freckles. He looks into the camera, completely deadpan, and says, "He comes in every Tuesday." The person filming wheezes with laughter. Dryers roaring, water dripping, a short bleat, traffic in the distance. Muted true colors with the yellow jacket as the one accent. No cars in the tunnel, no logos, no text.”

Direct-flash photo of a woman in a cobalt windbreaker laughing on the hood of a white car covered in pink foam in a self-serve car wash bay, made with MiniMax’s own image model as the start frame for a MiniMax H3 clip in Made

Start-frame control

“Vertical phone photo at night with harsh direct on-camera flash, inside a self-serve car wash bay. An attractive Black woman in her mid-20s sits on the hood of a small white hatchback covered in thick pink soap foam, leaning back on her hands and laughing at the camera. She wears an oversized cobalt-blue nylon windbreaker over a white tank top, black cargo pants and chunky white sneakers; her dark curly hair is up in a high bun with loose curls around her face. Clear, even skin with light film makeup, no freckles, her face the brightest thing in the frame. Wet concrete floor with puddles, a coiled pressure-wash hose and a foam brush hanging on the tiled wall behind her, fluorescent tubes overhead, pink foam dripping off the bumper. Candid and unposed, slight motion blur, muted true colors with the pink foam as the one accent. No text, no signs, no logos, no watermark.”

A frame from the MiniMax H3 car wash clip, the passenger laughing as pink foam and blue brushes cover the glass, generated in Made

Frame from the action clip

“A single frame from the clip above, straight from MiniMax H3.”

A frame from the MiniMax H3 sheep clip, the puffed-up sheep standing at the car wash exit, generated in Made

Frame from the 8 second scene

“A single frame from the clip above, straight from MiniMax H3.”

How to write a MiniMax H3 prompt

  1. 1

    Choose text or image input

    Write a prompt on its own, or add a start image to set the first frame. With a start image set, you can add an end frame too.

  2. 2

    Write what we hear

    Name the sounds, and put any line someone says in quotes. Every clip comes with sound.

  3. 3

    Set format and length

    Pick 21:9, 16:9, 4:3, 1:1, 3:4, or 9:16 for text-to-video, and any length from 5 to 15 seconds.

  4. 4

    Pick a resolution

    Clips start at 480P. 768P is the sharpest native render, and 2K and 4K upscale a 768P render.

What MiniMax H3 responds well to

Prompt for these jobs

  • Clips with sound: voices, laughter, and room noise come with every video
  • Talking clips: write a line in quotes and the person on screen says it
  • Animating a photo from a start frame, with an optional end frame for where the clip lands
  • 2K and 4K files for big screens, upscaled from a 768P render

Check the output for

  • Sound is always on, with no silent mode. Image-to-video follows the start image’s framing, so the aspect ratio setting applies to text-to-video only.
  • 480P and 768P are the native renders. 2K and 4K are upscales of a 768P render and cost more per second, so draft at 480P, the default.
  • It adds details you didn’t write. In our tests it put a made-up badge and plate on a car and a print on a jacket, and in a clip we wrote without dialogue, someone said a line anyway. Asking for a plain jacket, an empty tunnel, and no logos or text fixed the next take.

Iteration is cheap here: MiniMax H3 costs from 0.5 credits / second, so rewriting a prompt and rerunning it is a normal part of the workflow, not a budget decision. See the full MiniMax H3 pricing breakdown.

Questions, answered

MiniMax H3 prompts FAQ

Need a specific answer? Contact support

It is MiniMax’s open-weight base video model, with its weights on Hugging Face since August 2026. It turns a prompt, a start frame, or a start and an end frame into a 5 to 15 second clip with sound.

More tested prompts across all models live in the prompt library, and the full MiniMax H3 capability rundown is on the model page.