Available now

MiniMax H3 LoRA: video from references

By MiniMax

MiniMax H3 LoRA makes 5 to 15 second videos with sound using one to three LoRA files you trained, from text, keyframes or reference media.

Model ID
Maker
MiniMax

Input

Fields marked * are required. Review the price before generating.

Text prompt for video generation. Refer to reference assets by their modality and order in the reference lists: Image 1, Image 2, Video 1, Audio 1, and so on.

Example prompt. Edit it or write your own.

The duration of the video in seconds.

480P and 768P are native generation modes; 2K and 4K upscale a 768P base result.

The subject/style reference images, referenced in the prompt as Image 1, Image 2, and so on. Reference images, videos, and audio clips must add up to at most 12 files.

Uploading a file needs an account. Create account

JPEG, PNG, WebP · up to 25 MB · 0/9

The motion/reference video clips (2-15 seconds each, combined duration at most 15 seconds), referenced in the prompt as Video 1, Video 2, and so on. Reference images, videos, and audio clips must add up to at most 12 files.

Uploading a file needs an account. Create account

MP4, WebM · up to 100 MB · 0/3

The reference audio clips (2-15 seconds each, combined duration at most 15 seconds), referenced in the prompt as Audio 1, Audio 2, and so on. Images, videos, and audio can be provided individually or together. Reference images, videos, and audio clips must add up to at most 12 files.

Uploading a file needs an account. Create account

MP3, WAV · up to 50 MB · 0/3

List of LoRA adapters to apply to the transformer.

Additional settings

Random seed. A random seed is selected when omitted.

Uploading a file needs an account. Create account

MP3, WAV · up to 50 MB

Estimated total · 5 s, 2K$1.29

You sign up first. Your settings are kept.

About this model

Overview

Every mode needs at least one LoRA: a link to the trained weights and a scale from 0 to 4. Videos run 5 to 15 seconds at 480P or 768P, or at 2K and 4K upscaled from a 768P render. You can also pin your own audio clip as the soundtrack.

Start from a first frame, a last frame or both, from text alone, or from up to 9 reference images, 3 reference videos and 3 audio clips, called Image 1, Video 1, Audio 1 in the prompt.

What you can build

  • Clips that keep a trained character or product look across scenes
  • Brand videos in a style you trained once
  • Scenes that move a trained subject through reference motion
  • Videos cut to your own music or voice track

Prompting tips

  • Start with the LoRA scale at 1 and raise it only if the trained look does not come through.
  • Use the trigger words your LoRA was trained with in the prompt.
  • Draft at 480P or 768P, then render the prompt you like at 2K or 4K.

What to check

Watch the entire clip and compare motion, identity, transitions and audio with your inputs.

Inputs

  • Prompt (required): Text prompt for video generation. Refer to reference assets by their modality and order in the reference lists: Image 1, Image 2, Video 1, Audio 1, and so on.
  • Duration (optional): The duration of the video in seconds. From 5 to 15.
  • Resolution (optional): The resolution of the generated video. 480P and 768P are native generation modes; 2K and 4K upscale a 768P base result. Options: 480P, 768P, 2K, 4K.
  • Seed (optional): Random seed. A random seed is selected when omitted.
  • Prompt Expansion Mode (optional): How much effort to spend rewriting the prompt before generation. 'disabled' skips prompt expansion. 'fast' returns in about a second. 'balanced' picks per request. 'quality' spends up to ~30s on a richer prompt.
  • Aspect Ratio (optional): The aspect ratio of the generated video. Options: adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16.
  • Reference Images (optional): URLs of subject/style reference images, referenced in the prompt as Image 1, Image 2, and so on. Reference images, videos, and audio clips must add up to at most 12 files.
  • Reference Videos (optional): URLs of motion/reference video clips (2-15 seconds each, combined duration at most 15 seconds), referenced in the prompt as Video 1, Video 2, and so on. Reference images, videos, and audio clips must add up to at most 12 files.
  • Reference Audios (optional): URLs of reference audio clips (2-15 seconds each, combined duration at most 15 seconds), referenced in the prompt as Audio 1, Audio 2, and so on. Images, videos, and audio can be provided individually or together. Reference images, videos, and audio clips must add up to at most 12 files.
  • Target Audio (optional): Optional URL of an audio clip at least 2 seconds long (maximum 15 MB) to pin to the generated soundtrack. Cannot be combined with reference audio. Longer clips are trimmed to the generated video's length, keeping the beginning; shorter clips are padded with silence. The clip is resampled to 32 kHz stereo and replaces the output soundtrack without changing playback speed. Accepts an HTTP(S) URL or a base64 data URI.
  • Loras (required): List of LoRA adapters to apply to the transformer.

Output and limits

  • Length: 5 seconds to 15 seconds.
  • Resolutions: 480P, 768P, 2K, 4K.
  • Aspect ratios: adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16.
  • Readable text inside the video, especially small text, is unreliable.
  • Scenes with several interacting characters can lose consistency.
  • Longer clips can show more motion drift or scene changes.
  • Low-quality inputs cause glitches: heavy compression, collages, screenshots and text overlays.
  • Some prompts or outputs can be refused by safety checks.

Price guide

  • 480P: $0.0991 / second
  • 768P: $0.1189 / second
  • 2K: $0.2575 / second
  • 4K: $0.317 / second
  • The total for your settings is rounded up to the next cent.