Back to MiniMax H3 prompts
Video edit variant / Reference Generation

Selective voxel edit A

A precise edit mask can be described in semantic terms: change these objects and protect the rest.

Official manual translationTranslated from a MiniMax H3 manual example
Actual H3 output / 4s
Prompt -> result

What this output demonstrates

This short variant selectively transforms moving street objects while leaving people and architecture photographic.

Control pattern
A precise edit mask can be described in semantic terms: change these objects and protect the rest.
Check in your run
A global style request will voxelize the complete frame.
Official manual example / English teaching translation / source reviewed 2026-08-08Open this example in the H3 manual
Shot structure
0s2s4s
INPUTR2V
ACTION01
CAMERA01
AUDIOON
Working principle

A precise edit mask can be described in semantic terms: change these objects and protect the rest.

Watch for

A global style request will voxelize the complete frame.

Adapt this brief

Prompt draft

Keep the source buildings, pedestrians, and overall environment realistic. Convert only the trees and cars into 3D pixel or voxel blocks using Image 1 for style. Preserve their motion paths plus the real scene shadows and transmitted light. Use Video 2 only for the balance between real surroundings and selective pixel objects.

Replace the subject and reference roles first. Change camera, timing, and sound one variable at a time.