← MiniMax H3 prompt guide

Dialogue and native sound

Scenes where people talk, written so the model generates the voices, lip movement and room sound together with the picture.

The templateAll 12 clips

Anime horror-night gag with burned-in subtitles

Shot list · MiniMax H3 per the author · Oct 6, 2026

Six shots have exact cut times, a field of view and a static camera each; the Japanese lines are spoken, and the English subtitles are specified as image-only text with typeface, outline, a single line, a fixed position and a rule that each one leaves at the cut. It reverses the usual subtitle ban on the same principle: every piece of on-screen text is listed exactly, and shots without a line are told to carry none. Gag faces are written as concrete anime conventions (shrinking pupils, shade lines, souls leaving the mouth), with a closing rule that faces return to the normal design.

Dialogue and native soundPerformance and emotionAnimated styles and art direction

Read @MayorKingAI's prompt
2D anime | romantic comedy | cel-shaded, clean line art, dark blue night palette.
Multishot, 6 shots. Cuts only at the specified points.

CHARACTERS
HANA: wavy shoulder-length dark red hair, oversized gray cardigan with one button missing, a bandage on her index finger. 160 cm.
YUTO: neat black hair with a small bed-head tuft at the back, navy hoodie with a pilled sleeve, glasses pushed up on his forehead. 178 cm.

LOCATION
Dark living room at night. They share one sofa under a single fleece blanket. A TV off-frame in front of them is the only light source, flickering blue on their faces. A bowl of popcorn between them.

SUBTITLES
Spoken dialogue is in Japanese. English subtitles are burned into the image as on-screen text only — they are never spoken aloud. Style: white sans-serif, thin black outline, one single line, bottom center of frame, 8% from the bottom edge. Each subtitle appears with its line and disappears at the cut. Subtitle text is shown inside 【】 brackets below; the brackets themselves are not displayed.

Shot 1 (0–2.5s) — Medium front shot of both, FOV 47, static.
Hana sits upright, arms crossed, smug smile, chin raised. Yuto beside her grips a cushion with both hands, already tense.
Hana (spoken, Japanese): 「全然怖くないし。」
On-screen subtitle: 【I'm not scared at all.】
HARD CUT

Shot 2 (2.5–5s) — Close-up on Yuto, FOV 29, slow push-in ending on his eyes.
His pupils shrink to tiny dots, a vertical column of blue shade lines drops over his upper face, sweat drop on his temple. He swallows, the cushion squeezed flat. No subtitle in this shot.
HARD CUT

Shot 3 (5–8s) — Medium front shot, FOV 47, static.
A violent white flash from the TV and a loud sting. Both jump. Clear gag faces: mouths huge and wavy, eyes white, hair standing in spikes. The popcorn bowl flips and popcorn bursts upward like confetti. Two small white ghost-shaped souls slip out of their open mouths.
Hana and Yuto (spoken, Japanese): 「ぎゃああああっ!」
On-screen subtitle: 【AAAAAAH!】
HARD CUT

Shot 4 (8–10.5s) — Wide shot, FOV 63, static.
They are clutching each other tightly, trembling, their outlines drawn with shaky wobbling lines. Popcorn rains down on them. The souls drift back into their mouths. No subtitle in this shot.
HARD CUT

Shot 5 (10.5–13s) — Two-shot close-up, profile, FOV 29, static.
Silence. They slowly realize how close they are, faces a hand apart. Both blush deep red at the same time. A tiny comic sparkle pops between them. They let go and shuffle apart in short jerky movements. No subtitle in this shot.
HARD CUT

Shot 6 (13–15s) — Close-up on Hana, three-quarter front, FOV 29, static.
Normal detailed anime face: cheeks red, eyes sliding sideways, lips pouting. One popcorn kernel stuck in her hair.
Hana (spoken, Japanese): 「…今のはノーカン。」
On-screen subtitle: 【...That one doesn't count.】
She pulls the blanket up over her nose. End on her eyes peeking above the blanket.

LIGHTING
Only light source is the off-frame TV: flickering cool blue 8500K, with a single white flash in shot 3. Deep shadows in the room.

Sound: No score. Production audio only — faint TV dialogue murmur, a sharp horror sting, popcorn scattering, fabric rustle, double scream, then near silence and a soft blanket rustle. All voices in Japanese only.

Characters keep the same face, hair and wardrobe in every shot. Gag faces are clear, exaggerated super-deformed expressions, then return to the normal detailed anime face. Clean consistent line art, consistent lighting between shots. Subtitles stay in the same position, size and style in every shot, spelled exactly as written. The room contains only the sofa, the blanket and the popcorn bowl.

Original post ↗

What holds it together

Because H3 generates sound with the picture, a dialogue prompt is also a sound script. The strongest cases in the library do four things. They give each speaker a stable label and a voice described in words. They quote every line exactly, in its own language, inside the shot where it is spoken, with an action before it and a reaction after. They plan the cutting around the lines: a prop or two-shot to set positions, then singles and over-the-shoulders, then a held close-up on whoever has the last reaction. And they end with separate notes for ambience and score, very often no music at all, so the voices sit in a quiet room. Off-screen speakers are marked as off-screen and the on-screen character's silence is stated. What goes wrong: too many words for the running time, speakers attached to the wrong face, lines that turn into subtitles, and long scenes on slower setups. One author reported dropped and blended words when a short scene carried too many lines.

Write these in

  1. Each speaker: who they are, where they stand, and a voice described in words
  2. Every line quoted exactly, in its original language, inside the shot where it is spoken
  3. A small action before each line and a beat after it
  4. A shot plan with cut times: who is framed, at what size, from which side
  5. Off-screen voices marked as off-screen, with the visible character's lips closed
  6. Ambience listed separately, and music stated as present or absent
  7. No subtitles or other on-screen text

What breaks it

  • Filling every second with speech. Fewer, shorter lines with pauses read as acting; wall-to-wall lines blur.
  • Describing a voice only as 'warm' or 'deep'. Give age, pitch, texture and pace.
  • Attaching a read-aloud voice sample when you want natural delivery. It tends to make lines sound dubbed; describe the voice in words instead.
  • Forgetting to ban captions. Write 'no subtitles, no on-screen text' or the lines may be printed on screen.
  • Asking for subtitles in one word. If you want them, give the exact words, typeface, outline, position and timing, and say they are shown, never spoken.

The template

[DURATION]-second, [aspect], live-action dialogue scene, [period and place]. Light: [source and direction]. Native dialogue in [language]; lips match every syllable.
Speaker 1 (S1): [age, look, outfit], holds screen-[left/right]. Voice: [age, pitch, texture, pace, accent].
Speaker 2 (S2): [same fields], holds the other side. Keep this axis in every shot.
Shot 1, 0-[X]s: [prop close-up or two-shot]. [S1's small action]. S1 says: "[line]". [S2's reaction].
Shot 2, cut at [X]s: [S2 single, lens, chest-up]. [Action before the line]. S2 says: "[line]". A short pause.
Shot 3, cut at [Y]s: [S1 close-up]. S2 continues off screen: "[line]"; S1 listens with lips closed, [visible reaction].
Shot 4, cut at [Z]s: [closing two-shot]. [Last line or silence]. Hold [0.5-1] seconds, no fade.
Sound: [room tone and two or three physical sounds]. Music: [none / instruments and tempo]. No subtitles, no on-screen text.

Ours to use freely. Replace everything in brackets.

12 clips that use it

Each plays from the creator's own post on X. Our note sits under it; the prompt belongs to them.

A servant comforts a worried princess

h3-native · MiniMax H3 per the author · Aug 29, 2026

Written in H3's native dialogue form: each speaker has a label and a delivery note, every line is quoted in English inside dialogue tags, and four shots cut at fixed times from a medium close-up to a held two-shot. Small hand actions (adjusting a strap, holding a forearm) carry the relationship between lines, and soundscape and score are listed separately.

Dialogue and native soundPerformance and emotionReference-driven characters

Read @ou_zhen599's prompt
For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] Live-action, cinematic realism. Both women fully match <Picture 1>: same faces, hair, age, body proportions, dresses, palace interior, window light, bench, noble atmosphere, and class contrast. Left woman is the servant: dark-haired, standing, gentle, careful, attentive. Right woman is Princess Emma: blonde, seated, noble, fragile, deeply worried. IMAX 70mm film grain, Nolan-inspired cinematic realism, photorealistic natural palette with charcoal black, umber, dark ochre, ash gray, and pale amber interwoven. Medium close-up at elegant eye level, soft daylight from the right windows, subtle handheld breathing, slow physical push-in. The servant adjusts Emma’s shoulder strap and smooths the fabric at her upper arm while Emma looks down and lightly tightens the fabric in her lap. The servant (S1), soft, warm, respectful, says: <d>[English] Princess Emma... the king is on his way home.</d> [Shot 2] At 00:04.000, the camera cuts tighter and drifts from the servant’s hand toward Emma’s face. The servant keeps one hand lightly on Emma’s shoulder. Emma finally looks up, but the worry does not soften. The servant (S1) says: <d>[English] He will be here soon. Please don’t fear so much.</d> Princess Emma (S2), refined, quiet, emotionally strained, replies: <d>[English] I am trying.</d> [Shot 3] At 00:08.000, the camera cuts to a gentle moving two-shot. Emma rises from the bench and steps toward the window on the right. The servant follows immediately from the left and lightly holds Emma’s forearm in support. The camera softly follows, keeping both women clear and Emma emotionally central. The servant (S1) says: <d>[English] You must trust he will return safely.</d> Princess Emma (S2) replies with slightly rising urgency: <d>[English] I want to... but something feels wrong.</d> [Shot 4] At 00:12.000, the camera cuts to a medium shot with a slight pullback and lift, then holds. Emma stands near the window on the right facing outward. The servant stands just left and slightly behind, smoothing the back of Emma’s sleeve and leaving her hand there in quiet support. The servant (S1) says softly: <d>[English] Emma...</d> Princess Emma (S2) answers: <d>[English] I’m afraid.</d> Final feeling is shared waiting, not relief. Only these two women are visible. The servant remains the comforter. Princess Emma remains the emotional center.

overall_soundscape: Quiet palace room tone, faint fabric rustle, light footsteps on stone floor, soft air near the window, and a distant chandelier creak.

non_diegetic_music: Quiet piano and soft low strings at a very slow tempo, restrained beneath the room tone, with only a slight rise in the final seconds.

Original post ↗

A governess and her employer: period dialogue in English

Shot-by-shot timeline · MiniMax H3 per the author · Aug 29, 2026

The prompt is written in Chinese but quotes the dialogue in English and asks for native English speech, with each voice described by register and emotional curve. Four shots cut on line endings around a fixed axis, each with second-by-second facial notes, followed by a sound section that rules out music.

Dialogue and native soundPerformance and emotion

A missing cufflink over tea

Shot-by-shot timeline · MiniMax H3 per the author · Aug 29, 2026

The scene is built so the evidence is the listener's face: the prompt names the words that trigger a fast blink, a swallow and a held breath. It opens on the prop and keeps both women chest-up across a fixed axis; the author's post says the scene carried too many lines for its length on his local setup and some words were dropped.

Dialogue and native soundPerformance and emotion

A king and his minister: the walls or the people

Shot-by-shot timeline · MiniMax H3 per the author · Aug 29, 2026

Positions encode rank (the king to the right and higher, the minister to the left and lower), and each line is paired with the facial change that lands on a given word. The turn comes in the last two lines, there is no music, and the closing silence gets 0.4 seconds before a hard end.

Dialogue and native soundPerformance and emotion

A tavern pour and a talk with an unseen guest

h3-native · MiniMax H3 per the author · Aug 27, 2026

A prop rule is stated and then repeated at every point where it could fail: the small jug never touches or rests on the large barrel. The shots move from pouring to a walk toward the lens to an offered glass, and the background is kept blurred so the server remains the only readable person.

Dialogue and native soundPerformance and emotionReference-driven characters

Read @ou_zhen599's prompt
For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] Live-action, cinematic realism, Jones fully matches <Picture 1>: same face, hair, age, body proportions, tavern maid outfit, sensual but grounded tavern presence, medieval tavern interior, large serving barrel, small pouring jug in her right hand, beer glass, warm firelight, candles, smoky air, and rustic wooden atmosphere. IMAX 70mm film grain, Nolan-inspired cinematic realism, photorealistic natural palette with charcoal black, umber, dark ochre, ash gray, and pale amber interwoven. Medium close-up from a slightly low front-left angle near the barrel, Jones slightly right of center, large barrel and flowing beer foreground left. Subtle handheld breathing and a gentle physical push-in. Background stays heavily blurred and non-readable so Jones is the only clear visible person. Jones pours beer carefully with believable foam control. The small jug stays in her right hand and never touches, rests on, leans against, or gets placed on the large barrel. The nearby off-screen man with a mature low sexy voice (S2) says in an off-screen voiceover: <d>[English] So this is the place you were saving for me?</d> Jones (S1), mature, warm, sensual, relaxed, confident, playful, answers with a faint smile while still pouring: <d>[English] Mm. Best drink in the house.</d> [Shot 2] At 00:03.500, the camera cuts to a moving medium close-up follow shot. Jones finishes pouring, does not place the jug on the barrel, and either sets it only on the front wooden bar counter or another separate wooden surface in front of her, or keeps it with her briefly away from the barrel. She lifts the filled glass and walks toward the unseen man. The camera tracks backward and slightly sideways to stay physically close, with no zoom. Mid-walk, she briefly tucks loose hair behind her ear, then continues. S2 says in an off-screen voiceover: <d>[English] That confident, huh?</d> Jones replies with a knowing half-smile: <d>[English] You’ll see.</d> [Shot 3] At 00:07.500, the camera cuts to an intimate close-up as Jones approaches, slightly left of center, the glass rising into the lower foreground. The camera retreats slightly, then softly stops. Jones leans in and offers the drink into his space without fully releasing it at first. She says in a lower intimate tone: <d>[English] Drink this first... then maybe I’ll show you the better part of the night.</d> The off-screen man (S2), closer now, says in an off-screen voiceover: <d>[English] Jones... you always talk like that to strangers?</d> Jones gives a tiny breath of a laugh. [Shot 4] At 00:11.500, the camera cuts to a close-up resolving into a medium close-up with a subtle pan and shift as Jones passes slightly across frame and settles at the bar edge or a nearby wooden post. The glass is now in his space or softly set onto the wooden bar. If the small jug is placed down later, it goes only onto the front bar counter or another separate wooden surface in front of her, never onto the large barrel. She shifts her weight, turns back toward the unseen man, and says quietly: <d>[English] Stay for one more.</d> Final beat holds on warm tavern ambience, soft glass touch on wood, low fire crackle, and her off-camera gaze toward him. Only Jones is visible as a clear readable person for the full video. No fantasy effects, no modern props, no exaggerated flirting, no overacting, no music-video energy.

overall_soundscape: Warm tavern ambience with low room chatter, distant laughter, fireplace crackle, beer pouring, soft glass-on-wood contact, fabric movement, and light footsteps on wooden floor. Everything stays intimate, grounded, and natural.

non_diegetic_music: Quiet low strings and subtle plucked acoustic textures at a slow tempo, kept restrained beneath the room tone.

Original post ↗

Where it shows up alongside other effects

Motivational type with a voice described in words

script · MiniMax H3 per the author · Aug 29, 2026

The voice-over is quoted in full, and the voice is described rather than supplied: calm, deep and natural, restrained at first and firmer later, with short pauses and no trailer delivery. A pacing guide gives each line its own visual behaviour, so the animation's energy rises with the script.

Kinetic type and title sequencesDialogue and native sound

Read @aimikoda's prompt
Create a cinematic motivational kinetic typography video built around the following voiceover and on-screen text:

"You don’t need to feel ready.

Start anyway.

You’ll doubt yourself.
You’ll get tired.
You’ll want to stop.

Keep going.

You don’t need perfect timing.
You don’t need permission.

Take the next step.

Then take another.

Just don’t quit before it works."

Style: bold, minimal, emotionally powerful kinetic typography. Use only typography, punctuation, abstract graphic shapes, negative space and motion design. No characters, no literal environments, no illustrative scenes. Let the text carry the full emotional impact.

Voice: calm, deep and natural male voice, slightly serious and motivational. Start restrained, then gradually become more confident and firm. Use short pauses between lines. No shouting or dramatic trailer-style delivery.

Tone: inspiring, intense, grounded, modern and uplifting. Start with a restrained and uncertain mood, then gradually build momentum, confidence and emotional force.

Visual language: clean black background with high-contrast white typography, with selective accent color such as red or warm orange for emphasis. Use bold condensed sans-serifs for impact lines, with occasional lighter or narrower type for quieter moments. The typography should feel sharp, cinematic and dynamic.

Animation: begin slow and tense, with pauses and hesitation built into the motion. As the speech progresses, increase speed, scale and confidence. Use strong scale shifts, snap zooms, rhythmic cuts, staggered word reveals, tracking changes, line breaks, layered repetition, subtle distortion and impactful punctuation. Key phrases should hit hard and feel memorable.

Pacing guide:
- "You don’t need to feel ready." appears with hesitation, space and restraint.
- "Start anyway." lands abruptly and decisively.
- "You’ll doubt yourself. / You’ll get tired. / You’ll want to stop." appears in a rhythmic sequence, each line gaining more emotional pressure.
- "Keep going." should break through that pressure like a command.
- "You don’t need perfect timing. / You don’t need permission." should feel liberating and defiant.
- "Take the next step." should feel simple, direct and actionable.
- "Then take another." should visually continue the motion forward, creating a sense of steady progress.
- "Just don’t quit before it works." should be the final emotional climax and strongest visual statement.

End with the final line held clearly on screen for emphasis. The whole piece should feel like a premium social media motivational typography video, clean, modern and emotionally effective.

Original post ↗

Cantonese corridor comedy about fake kung fu

Shot-by-shot timeline · MiniMax H3 per the author · Aug 28, 2026

The comedy comes from pauses and props rather than action: a rattling cup lid, a toothpick test, a reverse shot in a mirror. The prompt asks for native Cantonese with the two speakers clearly separated, and bans full-body shots and real film characters.

Performance and emotionDialogue and native sound

Last words before an assassination

Shot-by-shot timeline · MiniMax H3 per the author · Aug 28, 2026

The prompt states what the scene is not (a fight) and moves the tension into a hand on a sword hilt that never draws and finally loosens half an inch. Each line has a facial cue tied to a word, and blocks of red, black and gold carry the mood in chest-up frames.

Performance and emotionDialogue and native sound

Police and thief, with the roles swapped

Shot-by-shot timeline · MiniMax H3 per the author · Aug 24, 2026

The author posted this as a failure. The two men are identified only as the left and right figures in the reference image, and the prompt then places them on the opposite sides of the frame; the post reports that officer and thief came back reversed, a clear argument for binding identity by appearance.

Reference-driven charactersDialogue and native sound

A talking barista from a nine-panel storyboard

storyboard · MiniMax H3 per the author · Aug 20, 2026

The storyboard handles the shots and the text handles the speech: four short lines in Japanese for her to say while making the latte. Each line matches a stage of the process, which puts the spoken script in the same order as the panels.

Reference-driven charactersDialogue and native sound

Alphabet lesson where letters become objects

Shot-by-shot timeline · MiniMax H3, inferred · Aug 3, 2026

Every letter follows the same five steps (letter, sound, object, action, word), and each morph maps part of the letter to part of the object, such as the point of the A becoming the apple's stem. Narrator lines, on-screen words and small sound effects sit in the same time block, and the closing list bans warped letters and duplicated objects.

Transformation and style shiftsKinetic type and title sequencesDialogue and native sound

Read @umesh_ai's prompt
Create a 15-second animated educational video that teaches young children the letters A, B, C, and D.

The learning pattern for every letter must be:

LETTER → SOUND → OBJECT → PLAYFUL ACTION → OBJECT NAME

Target audience: children ages 3 to 6.

Visual style:
Use adorable rounded 3D characters, soft pastel colors, gentle facial expressions, and simple recognizable objects. Combine this with a premium minimalist technology aesthetic featuring clean white space, elegant composition, soft studio lighting, subtle reflections, smooth gradients, rounded geometry, crisp typography, and extremely polished transitions.

The animation should feel playful and child-friendly while remaining calm, uncluttered, and beautifully designed.

Use a clean off-white background with a different soft color glow behind each letter.

0:00–0:01 | Introduction
A small smiling star mascot bounces into the center of the screen.

Colorful letters briefly float around it.

Display the text:

“Let’s learn!”

The mascot taps the screen, creating a soft ripple that reveals the first letter.

0:01–0:04 | A is for Apple

Show a large uppercase “A” and smaller lowercase “a” beside it.
Use thick, rounded, highly readable typography.

The narrator says:

“A. A says ah. A is for Apple.”

The uppercase A gently inflates and transforms into a shiny red apple.

Its top point becomes the apple stem, and a small green leaf unfolds from the side.

The apple gains a cute smiling face and performs one soft bounce.

Display the word:

“APPLE”

Highlight the first letter A in red.

Add a soft pop and a tiny crunchy sound.

0:04–0:07 | B is for Ball

The apple rolls across the screen and leaves behind a curved red trail.

The trail loops twice and forms a large uppercase “B,” with a lowercase “b” appearing beside it.
The narrator says:

“B. B says buh. B is for Ball.”

The two rounded sections of the B expand and merge into a colorful striped ball.

The ball bounces twice with playful squash-and-stretch animation.

Display the word:

“BALL”

Highlight the first letter B in blue.

Synchronize each bounce with a soft musical note.

0:07–0:10 | C is for Cat

On its final bounce, the ball stretches into a curved shape and becomes a large uppercase “C.”

A lowercase “c” slides gently into place beside it.

The narrator says:
“C. C says kuh. C is for Cat.”

The C rotates and becomes the curled tail of a cute orange cat.

The rest of the cat forms from soft rounded shapes.

The cat stretches, blinks, and gives one gentle wave with its paw.

Display the word:

“CAT”

Highlight the first letter C in orange.

Add a quiet and friendly “meow.”

0:10–0:13 | D is for Duck

The cat’s tail uncurls and transforms into the curved side of a large uppercase “D.”

A lowercase “d” pops up beside it.

The narrator says:

“D. D says duh. D is for Duck.”
The straight line of the D becomes the duck’s neck.

The curved section becomes its round yellow body.

A small orange beak and two tiny wings pop into place.

The duck waddles forward, flaps its wings, and gives one cheerful quack.

Display the word:

“DUCK”

Highlight the first letter D in yellow.

Add tiny water ripples beneath its feet.

0:13–0:15 | Recap

The apple, ball, cat, and duck slide into four clean rounded tiles.

Place their letters above them:

“A  B  C  D”

The mascot returns and points to each object as they bounce once in sequence.

Narrator:
“A, B, C, D. Great job!”

Finish with the text:

“Great job!”

Use a small sparkle animation and a warm musical chime.

Animation requirements:

Keep each letter fully visible for a moment before it transforms.

Show uppercase and lowercase versions clearly.

Make every object instantly recognizable.

Use smooth shape morphing so children can visually understand how the letter becomes the object.

Maintain stable spelling, clean letterforms, accurate object shapes, and consistent character design.

Use gentle squash-and-stretch, soft motion blur, subtle shadows, polished lighting, and precisely synchronized sound effects.
Avoid fast camera movement, cluttered backgrounds, harsh colors, tiny text, warped letters, random symbols, duplicated objects, scary expressions, or overly complex transformations.

The final video should feel cute, educational, memorable, calming, and exceptionally polished.

Original post ↗

The week's best new dramas and AI movies, in one email.

We are putting together a weekly email: the new dramas and AI movies worth your time, and any free-coin deals we find. Leave your address and you will get the first issue when it starts. No spam.