Prompt Library & Guide
A curated library of MiniMax H3 prompts for text to video, image to video, and reference video generation. Learn the five-layer MiniMax H3 prompt formula, master the prompting techniques the model responds to best, and copy any example straight into the generator.
Share this prompt library
A strong MiniMax H3 prompt stacks five layers in one pass. MiniMax H3 accepts prompts of up to 7,000 characters — long enough for a full shot list with sound design — so be specific at every layer.
Who or what is in the shot, and what happens. Lock identity with concrete details — "a woman with a short bleached haircut and a red satin dress" beats "a stylish woman".
e.g. A barista with tattooed forearms pulls an espresso shot
H3 reads cinematography vocabulary directly. Name the shot size, lens, and movement: close-up on a 85mm, slow dolly-in, handheld with subtle shake, high-angle establishing shot.
e.g. macro close-up, slow lateral dolly, shallow depth of field
Define the look: golden hour backlight, neon rim light, 35mm film grain, Wes Anderson symmetry. A named visual language keeps all shots in one clip consistent.
e.g. warm morning window light, fine film grain, muted palette
H3 generates native stereo audio, so direct the sound like a shot: room tone, foley details, where the music lands. Audio direction is what separates great H3 prompts from average ones.
e.g. espresso machine hiss, cup clink on saucer, soft jazz bass
State what you do not want. Negative direction is unusually effective with MiniMax H3: no garbled text, no identity drift, no jump scares, no soft dissolves.
e.g. no on-screen text, no extra customers, no oversaturated colors
Here is what the five layers look like assembled into one ready-to-run prompt. Notice how every reference gets an explicit job, the camera and audio are directed separately, and the constraints close the prompt.
Cinematic product film. Use Image 1 as the product reference and preserve its design and label exactly. Camera: slow 360-degree orbit on a black reflective studio floor, one macro pass across the surface details. Lighting & style: soft top light with two cool rim lights, premium commercial look, shallow depth of field. Audio: quiet airy whoosh, minimal electronic bed, one restrained hit as the product settles into the hero angle. Constraints: no extra props, no warped logos or garbled text.
These techniques come straight from how the MiniMax H3 model is built to read prompts. Apply them to any example in the library below.
Don't just attach references — tell H3 what each one is for: "Use Image 1 for mood and location, Image 2 for the talent, Image 3 for the product." Explicit jobs beat unlabelled inputs every time.
For anything longer than one beat, storyboard inside the prompt: "[0–4s] wide establishing shot… [4–9s] push in on the product… [9–15s] hero angle." H3 follows the structure and the pacing holds.
Audio is generated natively, so art-direct it: "ice tapping crystal, faint cigar burn, controlled breathing." For music, describe instrumentation and where the beat lands instead of naming a song.
Negative direction is unusually effective with MiniMax H3: "no soft dissolves," "no garbled text," "no jump scares." Specific exclusions keep a stylized prompt from sliding into a nearby genre.
When a character must survive the whole shot, list the defining details: "preserve the short bleached haircut, red satin dress, and silver hoop earrings." Naming features gives the model something to hold onto.
Editing prompts work best as explicit substitution lists: "replace the newspaper with a green hardcover book; replace the chair with a red sofa." Always add what must stay unchanged to get a localized edit, not a re-generated shot.
H3 translates cinematography vocabulary directly: "rack focus," "wide angle with strong perspective distortion," "fine grain and soft highlight halation." Speak like a director and it shoots like one.
Cuts land better written as physical actions than named effects: "whip pan with motion blur, cut at peak blur, snap back into focus" beats "fast transition."
24 prompts found
Use Image 1 as the product reference and preserve its shape, label text, and finish exactly as shown — do not redesign or reinterpret it. Continuous 360-degree orbit around the product standing on a black reflective studio floor. Soft top light with two cool rim lights; a slow macro pass across the surface details mid-turn. End on a centered hero angle as a soft spotlight tightens. Audio: a quiet airy whoosh and a minimal electronic bed. No extra props, no garbled text, no warped logos.
Use Image 1 for the product and Image 2 for the setting. Present the product in a bright minimalist lifestyle set: [0–4s] locked-off wide shot as morning light slides across the set; [4–9s] push in while a hand enters frame and demonstrates the product's key feature; [9–15s] macro close-up of the detail, then settle on the product centered with a clean background. Keep the product design and label exactly as the reference. Audio: upbeat corporate track, soft clicks on each interaction. No on-screen text, no hands covering the brand.
Ultra-slow-motion commercial macro shot: ice-cold condensation beads rolling down a glass bottle, then a slice of fresh fruit dropping into fizzy liquid. 1000fps feel, backlit droplets, shallow depth of field, dark moody background with one warm key light. Camera: slow lateral dolly with a subtle rack focus from droplets to label. Audio: crisp fizz, a deep liquid splash, restrained bass hit. No plastic-looking textures, no artificial oversaturation.
Use Image 1 for mood, location, and film texture; Image 2 for the talent; Image 3 for the featured garment. A 15-second fashion film: the talent walks through a wind-swept desert highway rest stop at golden hour, the garment moving naturally with each step. Medium tracking shot, then a low-angle hero pass, ending on a static profile as the wind settles. Photoreal 35mm film look with fine grain and soft halation. Audio: wind, distant traffic hum, a cool minimal beat. Do not change the talent's face, hair, or the garment's design.
Epic theatrical sci-fi teaser. Fast, hard cuts only: a shaking command deck, white-hot hull flashes, split-second black frames, and a violent jump-to-warp impact. Volumetric light, anamorphic flares, deep shadows. A title emerges from deep-space shadow and resolves from soft blur into sharp focus. Audio: deep sub-bass pulse, rising strings, one restrained hit as the title locks. No slow dissolves, no cheap lens flares covering faces, no garbled title text.
Premium wildlife documentary sequence: a golden eagle glides low over a misty highland lake at dawn, talons skimming the water surface. Long-lens compression, slow tracking pan following the bird, then a cut to a macro of water droplets on feathers. Cold blue palette with a warm rim from the rising sun, atmospheric haze, photoreal detail. Audio: wind, wingbeats, distant water, a restrained orchestral pad. No anthropomorphic behavior, no oversaturated colors.
Black-and-white film noir opening. Night, rain-slicked city street seen through venetian-blind shadows. A lone figure in a trench coat walks toward the camera under a flickering streetlamp; smoke drifts through the key light. Slow push-in, high contrast, heavy grain, hard shadows. Audio: rain, distant traffic, slow jazz bass with brush drums, a single thunderclap. No modern signage, no color accents, no fast camera moves.
Sweeping historical epic establishing shot: a massive stone citadel at golden hour, banners rippling in the wind as thousands of torches flicker along the ramparts. Slow aerial crane rise revealing the valley and an advancing army below. Dust in the air, warm low sun, deep shadows, photoreal period detail. Audio: wind over stone, distant drums, a swelling choir. No modern elements, no fantasy creatures, no cartoonish crowds.
9:16 vertical short-drama scene. Use Image 1 and Image 2 as strict identity references for the two leads and preserve their faces, hair, and wardrobe throughout. A rooftop confrontation at dusk: one lead reveals a secret, the other steps back in disbelief, then grabs their wrist as they turn to leave. Medium-close shots with tight shot/reverse-shot cutting; grounded, natural acting — never theatrical. Audio: city ambience, wind, two clear dialogue lines. No subtitles, no watermark, no exaggerated expressions.
Use Image 1 as the presenter reference, preserving their face and hairstyle exactly. 9:16 selfie-style UGC ad: the presenter holds the product up beside their face, speaks directly to camera with natural enthusiasm, then tilts the product to show the label. Slight handheld shake, window-lit home interior, authentic phone-camera feel. Audio: clear natural speech, room tone, subtle upbeat bed. No studio lighting, no on-screen text overlays, no distorted product label.
Animate the uploaded image into a punchy meme reaction: the subject slowly leans toward the camera and does a double-take with widening eyes, holds one beat, then breaks into laughter. Keep the subject's face, hairstyle, and clothing identical to the input image. Slight zoom-in on the reaction beat, punchy timing, meme-friendly pacing. Audio: a record scratch, then a short comedic sting. No distortion of facial features, no extra characters.
Use Image 1 as the performer reference, preserving identity, hairstyle, and outfit exactly. Make the performer sing along to Audio 1, matching the vocal phrasing, mouth shapes, and stage presence — expressive but natural, with small gestures on the accents. Medium shot with a slow push-in, stage lights in a soft bokeh background. Audio: follow Audio 1 closely, keep it clean. No uncanny lip movement, no identity drift, no background crowd.
Abstract music visualizer: liquid chrome ribbons and neon particles pulse and morph in perfect sync with a driving electronic beat. Black void background, high-gloss reflections, rhythmic camera drifts that land on each downbeat. Transitions happen as physical events — ribbons snap, shatter into particles, and re-form on the drop. Audio: driving electronic track with a clear build, drop, and outro. No random flicker, no drift off the beat, no text.
Animate the uploaded poster while preserving its layout, typography, and color palette exactly. Layered parallax: the background drifts slowly, mid-ground elements float, the headline typography subtly tracks wider. Light sweeps across the surface on the beat, and a soft type-on click sounds whenever new text elements move. Shallow simulated depth, cinematic pacing. No rearranged layout, no distorted letters, no added imagery.
Use Image 1 for the graphic style and color system. A 15-second opening title sequence built as motion-graphics collage: linework draws on over black, split-screen boundaries snap into place, credits slide in through masks on the beat. Bold geometric color fields, asymmetric frames, crisp rhythmic transitions — no soft dissolves. All credit text must stay clean and legible. Audio: an original suspense-jazz cue; low drone and pizzicato first, sparse drums entering at 6 seconds, a tense chord to finish. No garbled characters, no repeated names.
Animate the uploaded logo into a premium ident. The logo assembles from fine particles of light that converge and lock into place; a subtle lens flare sweeps across the surface, then everything settles into a still, confident hero frame. Dark gradient background, precise mechanical easing, restrained glow. Audio: a rising shimmer, one deep impact hit as the logo locks, then silence. No bouncing, no spinning, no warped logo geometry.
Stop-motion claymation. A small clay fox sprints to the edge of a clay cliff and leaps in dramatic slow motion over a bubbling lava canyon, limbs fully stretched, clay fingerprints visible on its body. Handcrafted texture, visible thumb marks, slightly uneven frame feel like 12fps stop motion. Camera races beneath the fox mid-air, revealing the canyon depth. Audio: springy footsteps, a held breath, a triumphant orchestral swell. No smooth CG surfaces, no photorealism.
High-budget anime action cut: a swordswoman dashes through a bamboo forest in the rain, one decisive slash in extreme slow motion with water droplets suspended mid-air, then snap back to full speed as bamboo culms fall around her. Dynamic perspective, speed lines, impact frames, cel shading with painterly backgrounds. Audio: rain, a blade ring-out, a taiko hit on the slash. No CGI look, no motion smear artifacts, no off-model faces.
Blend live-action footage with hand-drawn luminous animation. A real kitchen at dusk, filmed one-handed on a phone with subtle shake and backlit exposure breathing; small glowing doodle creatures peek out from behind the mugs and wave slowly, curiously. The drawn elements should feel sketchy, warm, and alive, interacting with the real light. Audio: room tone, a soft mug clink, gentle electronic chirps from the creatures. No jump scares, no horror elements, no polished commercial look.
Use Image 1 as a strict character reference — preserve the face, proportions, costume design, and material detail exactly. A polished 3D animation showcase: the character steps into frame, turns playfully toward the camera, and strikes a signature pose as soft studio light sweeps around them. High-end feature-animation rendering, subsurface skin, fabric micro-detail, shallow depth of field. Audio: light footstep foley, a warm orchestral flourish on the pose. No stylization drift, no changed costume, no uncanny expressions.
Remove the green-screen background from Video 1 and replace it with a fairy-tale forest environment similar to Video 2. Every background element must respond correctly to the subject's movement, with accurate parallax and contact shadows. Relight the subject so rim light, color temperature, and reflections fully match the new scene. Keep the subject's performance, timing, and identity unchanged. No green edges, no spillover, no compositing seams.
Relight the scene in Video 1 from daytime to night. Replace the sky with deep blue tones and faint stars, drop the overall exposure, convert practical windows and street lamps into warm glowing sources, and add soft moonlight rim on the subject. Preserve the original camera motion, subject performance, and composition frame by frame. Audio: shift the ambience to crickets and a distant night wind. No flickering exposure, no ghosting, no changed motion.
In Video 1, replace the drink can the person is holding with the product shown in Image 1, keeping the new product's label crisp, correctly oriented, and readable in every frame. Match the hand's grip, the condensation, and the lighting on the new can to the original scene. Everything else in the video stays exactly the same — same motion, same background, same person. No warped label text, no hand geometry changes, no flicker on the swap.
Add warm orange hand-drawn marks in the style of Image 1 around the two people in Video 1. As they walk closer, the marks multiply and build from tiny sparks into bright radiance around them; when they finally high-five, a burst of playful doodle stars and confetti pops. The drawn elements should follow the characters' motion with lively, sketchy energy. Keep the original footage untouched otherwise. Audio: pencil-scratch textures, a pop on the high-five. No covering faces, no photoreal VFX look.
