MiniMax H3 prompt library

MiniMax H3 Prompts: 9 Shots, 9 Styles, Copy-Ready

Every line here is a full H3 prompt — style tag, shot block, dialogue, sound — and the frame beside it is what that line rendered.

Raiden · the blade — The same render, cut in close: the arc leaving the scabbard is what the low camera was for.

Gojo · the blindfold — The blindfold comes down on the beat; the neon behind him is the only light left in the frame.

Hori · mid-line — The half second she answers the camera. In the documentary style, that is the acting.

Homelander · the smile — Cropped from the CCTV tile: the smile he gives the lens just before the tape tears.

Anime cel · Raiden Shogun — The blade clears the scabbard, lightning climbs the steel, the camera pushes in. The style tag, the low angle and her one spoken word are all in the line. — **Anime cel · Raiden Shogun** — The blade clears the scabbard, lightning climbs the steel, the camera pushes in. The style tag, the low angle and her one spoken word are all in the line.

Aerial · Zhongli — A drone falls past lit roofs to find one man on a rooftop. The prompt fixes the height, the drift, and the second the gold spear forms. — **Aerial · Zhongli** — A drone falls past lit roofs to find one man on a rooftop. The prompt fixes the height, the drift, and the second the gold spear forms.

Luxury spot · Gojo Satoru — Anamorphic, one hard key light, a five percent push in. The line says where the flare goes and when he pulls the blindfold down. — **Luxury spot · Gojo Satoru** — Anamorphic, one hard key light, a five percent push in. The line says where the flare goes and when he pulls the blindfold down.

Phone footage · Megumi Fushiguro — Shadows pour out of the ground into two heads with pale eyes. The line asks for a bad handheld camera on purpose — that is the style. — **Phone footage · Megumi Fushiguro** — Shadows pour out of the ground into two heads with pale eyes. The line asks for a bad handheld camera on purpose — that is the style.

VHS tape · Nobara Kugisaki — The hammer comes over the top and the papers go sideways. The line names the decade and the tape damage; the flinch lands in the clip. — **VHS tape · Nobara Kugisaki** — The hammer comes over the top and the papers go sideways. The line names the decade and the tape damage; the flinch lands in the clip.

Documentary · Hori Kyouko — Available light, no grade, one line said straight to the lens. The prompt sets how far away she is and who she is answering. — **Documentary · Hori Kyouko** — Available light, no grade, one line said straight to the lens. The prompt sets how far away she is and who she is answering.

16mm film · Miyamura Izumi — Warm halation on a dusk gate. The rack focus from fence to face is written down, not left to the model. — **16mm film · Miyamura Izumi** — Warm halation on a dusk gate. The rack focus from fence to face is written down, not left to the model.

CCTV · Homelander — A fisheye corner angle and a corridor with no exit. The prompt states the camera height, the wave, and when the picture tears. — **CCTV · Homelander** — A fisheye corner angle and a corridor with no exit. The prompt states the camera height, the wave, and when the picture tears.

Stylized CG · Starlight — Subsurface skin, a ten degree orbit, light gathering in her palms. The render look is set before the action starts. — **Stylized CG · Starlight** — Subsurface skin, a ten degree orbit, light gathering in her palms. The render look is set before the action starts.

The two-shot · 10 seconds — One prompt, two shots: the gate, a cut at 00:04.000, then the walk up. Both shots carry dialogue, and the sound and music lines run underneath.

The exact prompt that made it +

[Shot 1] Anime, cinematic cel animation, late-afternoon summer light, warm lens flare. A school gate: the girl in the first reference image waits under the tree with her bag held in front of her, looks up the path, then smiles and lifts one hand. Camera at her eye level, static, the leaves moving in front of the lens. She says: <d>[Japanese] 遅い。</d>
[Shot 2] At 00:04.000 the shot cuts to the boy in the second reference image walking up the path in his uniform, hair in his eyes; he raises a hand in apology and says: <d>[Japanese] ごめん、掃除当番。</d> Camera tracks beside him at walking pace, then settles as he reaches her and the two of them fall into step toward the gate.
Sound: cicadas, footsteps on gravel, a bag strap shifting, both voices close and natural.
Music: a warm piano figure with light strings underneath.

Where a prompt goes after you copy it

Neta Studio is a world you build by talking to it. The clip lands there as a work with its prompt saved beside it, so the same cast can be re-shot in another style without describing them again — and the line stays editable after the render.

One cast, nine styles. Characters live in the world, so the ninth render still has the same face as the first.
The prompt is part of the output. Every render comes back with the exact line that made it, ready to change one word and shoot again.
Shot language, not luck. Camera move, light and cut time are written in the prompt, so a re-shoot is a rewrite, not a reroll.

No H3 prompt skill in the directory yet — the button sends the format rules with your subject: style line, shot block, dialogue, sound.

Questions about H3 prompts

What does a working MiniMax H3 prompt look like?

Five parts, in this order: a style line naming the medium and the look, a [Shot 1] block with place, time, light and one camera move, the dialogue as <d>[language] line</d>, a Sound line, a Music line. The nine lines on this page are complete examples, not fragments.

Why do the shots have timecodes?

H3 reads shot blocks, and a cut written as 00:04.000 tells it when to change picture. Without the timecode the model decides for you, which is how a two-shot turns into one long take.

How is dialogue written?

Name the speaker and their voice first, then wrap the words in a dialogue tag: <d>[Japanese] 退け。</d> H3 generates native audio, so the line is spoken in the clip rather than added later.

What are the Sound and Music lines for?

Sound covers what is in the room — rain, footsteps, tape hiss. Music is the score, and writing Music: none is a real instruction when you want the clip to stay dry.

How long should a prompt and a clip be?

One shot, five seconds, under 150 words of scene description works best. A two-shot wants about ten seconds and one cut around the four-second mark. H3 goes up to fifteen seconds, but past ten a single line of action starts repeating itself.

Will the character stay on model?

Yes, when the character comes from the world you are shooting in. The reference image carries the face; the prompt only mentions wardrobe where it changes. A character you merely describe drifts between takes, and a two-shot with two references can pull the render style apart — check the frames before you spend a longer clip.

Can I use my own subject instead of a franchise character?

Yes. Add the photo to the world first so it can be used as the reference, then name it in the prompt. If there is no reference, describe the face in the same words in every prompt — consistency comes from repeating the description, not from wishing.

What do I get back?

The rendered clip and the exact prompt that produced it. Nothing else is rewritten, so the line you copy is the line that ran.

Nine lines. Yours to rewrite.

Keep the format, change the subject, shoot it again.

Make My Clip