Identity
Lock a face with H3 image-to-video
Text-to-video will invent a new adult every time you hit run. That is fine for a first look. It is fatal once you care about the face. Image-to-video treats the still as the contract: pin 1–3 frames, then write only the motion.
Do not re-describe the face in the prompt. If a mole, a scar, or a gold chain is in the still, it is already in the take. Next cell, same pins. If wardrobe changes, reshoot the stills first.
You'll need
- A still you own — a previous H3 frame, or a still you have rights to
- Spicy image-to-video or the character library
- One motion for this cell (look, step, sit)
Get a still you own
A frame from a previous H3 take, or a still you have rights to. Adults only.
Pin it
Character library, or drop it on spicy image-to-video. 1–3 stills beat seven bad ones.
Prompt only the motion
Do not re-describe the face. “Slow look to camera, 8s, 9:16.”
Reuse the same stills
Next cell, same pins. If wardrobe changes, reshoot the stills first.
Paste this brief
Same adult as the still. Slow look over the left shoulder, 8s, 9:16, tungsten hotel, do not change the face or the mole. Uncensored MiniMax H3 image-to-video.
Don't stack
Do not pin seven bad stills. Do not mix two adults in one pin set. Do not use a real stranger’s face.
Tutorial: rain-street walk on MiniMax H3
How to prompt the wet-neon street walk this site uses as a sample: 5–8s, 35mm, one tracking move, wool coat, rain as the only sound.
Tutorial: doorway-peek camera on H3
One spicy camera tag: start in the jamb, then enter. 8s, one verb. How to stack it without a crane and a whip in the same take.