Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/nidhinjs/prompt-master/llms.txt

Use this file to discover all available pages before exploring further.

Video, voice, and 3D AI tools sit at the edges of what most prompt guides cover — yet they have some of the most distinct and unforgiving syntax requirements. Sora needs film direction language, not scene description prose. ElevenLabs ignores adjective-laden descriptions entirely and responds to emotion markers and SSML-like pacing controls. ComfyUI-style 3D tools like Meshy and Rodin need export format specifications alongside the visual description. Prompt Master routes each of these tools through purpose-built frameworks drawn from how their underlying models actually process input.

Sora

Sora generates video by interpreting prompts as film direction instructions. The camera movement specification is not decorative — it is a primary input that shapes the entire shot. A prompt that describes a scene without a camera directive will produce an unpredictable default movement.
  • Direct the shot, not just the scene — write as a film director giving shot instructions, not as an image prompt
  • Camera movement is critical — always specify: static, slow dolly in, crane shot rising, tracking shot left to right, handheld, aerial pullback
  • Describe duration and pacing5-second shot, slow motion, real-time, time-lapse over 10 seconds
  • Include lighting, time of day, and weather as production context — these are treated as controllable production variables, not scene flavor
Cinematic tracking shot following a lone wolf running through a dense pine forest at dusk.
Camera: low-angle tracking shot at ground level, moving with the wolf at constant speed.
Lighting: golden hour, shafts of light breaking through the trees, long shadows.
Motion: wolf running at full speed, snow kicking up with each stride.
Duration: 6-second continuous shot, no cuts.
Style: photorealistic, 4K, film grain, anamorphic lens feel.

Camera Movement Reference

MovementDescriptionUse Case
staticCamera does not movePortraits, dialogue, still subjects
slow dolly inGradual forward pushBuilding tension, reveal shots
crane shot risingCamera lifts upwardEpic reveals, environmental scale
tracking shotFollows a moving subjectAction, running, vehicles
handheldSubtle organic shakeDocumentary feel, intimacy
aerial pullbackPulls back and up from aboveEstablishing scale, endings
pan left/rightHorizontal rotation on axisReveals, following a gaze

Runway Gen-3

Runway Gen-3 responds well to cinematic production language — the vocabulary of film production rather than consumer photography. Reference film styles and cinematographers to anchor the visual language.
  • Use cinematic languagewide establishing shot, close-up on hands, medium two-shot, rack focus from foreground to background
  • Reference film styles as anchorsin the visual style of Denis Villeneuve, Blade Runner color palette, Wes Anderson symmetrical composition
  • Describe color grading explicitlydesaturated with teal and orange grade, high contrast black and white, warm golden tones
  • Specify lens and depth of fieldanamorphic lens, wide angle with distortion, telephoto compression, deep focus
Wide establishing shot of a brutalist government building at night.
Visual reference: early Christopher Nolan — deep focus, cold blue-grey tones.
Camera: static wide shot, symmetrical composition, slight upward tilt.
Lighting: harsh sodium vapor streetlights casting amber pools, deep shadows on the building facade.
Color grade: desaturated, cold, teal shadows with amber highlights.
Lens: anamorphic, slight lens flare from the nearest streetlight.
Duration: 4-second static hold, subtle camera breathing.

Kling

Kling is a video AI model with particular strength in realistic human and physical motion. When the subject involves a person moving, body movement description is the highest-priority input — vague motion descriptions produce generic, unconvincing movement.
  • Describe body movement explicitlyslowly raises both arms above her head, turns to look over her left shoulder, hair sweeping across face, walks with a slight limp, favoring the right leg
  • Strong at realistic human motion — prioritize detailed movement descriptions for any human subject
  • Physical interaction details — describe how subjects interact with objects and environment: fingers trailing along the surface of the wall, feet splashing through shallow puddles
  • Combine with environmental motionwind moving through grass, rain falling at a 30-degree angle
Medium shot of a ballet dancer in a white studio.
Motion: She begins with arms at her sides, slowly raises both arms into fifth position above her head, then transitions into a slow pirouette — one full rotation, controlled, arms extended.
Body: Tall posture, pointed toes throughout, slight lean back at the peak of the arm raise.
Environment: Studio is bare, white walls, hardwood floor reflecting soft diffused window light from the left.
Camera: Static medium shot, dancer centered in frame.
Duration: 8 seconds, real-time speed.

LTX Video

LTX Video is a fast, prompt-sensitive video model. It responds to concise, visual, high-signal descriptions — elaborate prose reduces coherence rather than improving it. Always specify resolution and motion intensity.
  • Concise and visual — keep descriptions tight; every word should be a distinct visual signal
  • Specify resolution720p, 1080p, 1920x1080
  • Specify motion intensitylow motion (slow, subtle), medium motion (natural), high motion (fast-paced, energetic)
  • Fast iteration tool — LTX Video’s speed makes it well-suited for quick concept testing before committing to a slower model
Aerial view over a dense city at night, rain falling, neon signs reflecting on wet streets below.
Resolution: 1080p
Motion: low motion — subtle rain movement, slow camera drift downward
Duration: 5 seconds
Style: cinematic, photorealistic

Dream Machine (Luma)

Dream Machine by Luma AI produces high-quality video with strong handling of lighting and lens characteristics. Referencing specific lighting setups, lens types, and color grading approaches produces more controlled results than generic descriptions.
  • Reference lighting setupsthree-point lighting, Rembrandt lighting, rim lighting from behind, practical lighting only (candles and lamps)
  • Specify lens type50mm prime, anamorphic widescreen, fisheye, telephoto 200mm
  • Color grading stylesbleach bypass, cross-processed, Kodak Portra film emulation, high-key low contrast
  • Combine into a production package — each of these elements works additively; specifying all three (lighting + lens + grade) gives the model precise anchors
Interior of a jazz bar late at night, one musician playing saxophone on a small stage.
Lighting: Rembrandt lighting from stage left, practical amber bulbs above the bar in the background, heavy shadow in the foreground.
Lens: 85mm prime, wide open aperture, shallow depth of field — musician sharp, background bokeh.
Color grade: warm, slightly desaturated, Kodak Portra 400 film emulation.
Camera: Slow dolly in toward the musician, starting from mid-room.
Duration: 7 seconds.

ElevenLabs

ElevenLabs generates speech audio from text. The critical distinction from every other AI tool in this guide is that descriptive prose does not work — writing adjectives about how you want the voice to sound has no effect. ElevenLabs responds to SSML-like inline markers, explicit emotion tags, and pacing controls embedded in or alongside the speech text.
Prose descriptions of voice quality don’t affect ElevenLabs output. Writing “speak in a calm, warm, reassuring tone” in a system prompt or instruction field has minimal effect. You must use inline markers, pauses, and emotion specifications to control delivery.
  • Emotion tag — specify the emotional register: [excited], [somber], [warm and reassuring], [urgent], [sarcastic]
  • Pacing[slow], [fast], [deliberate pause before this word]
  • Emphasis markers — use *word* or [emphasis]word[/emphasis] for stressed syllables
  • Speech rate — specify words-per-minute or relative rate: speech rate: 0.85 (slower), speech rate: 1.1 (slightly faster)
  • SSML-like markers<break time="1s"/> for pauses, <emphasis level="strong">word</emphasis> for stress
Emotion: warm, sincere, slightly reflective
Speech rate: 0.9 (slightly slower than normal)
Emphasis: place emphasis on "every" and "you"

Text:
"We built this for <emphasis>every</emphasis> person who ever felt like the tools weren't made for <emphasis>them</emphasis>. <break time="0.8s"/> That changes today."
For long-form narration, break the script into emotional segments and specify the emotion at the start of each segment. A single emotion tag for a 5-minute script will drift — re-anchor every 30–60 seconds.

3D AI — Meshy / Tripo / Rodin

Meshy, Tripo, and Rodin generate 3D models from text prompts. They share a common prompt structure that combines visual description with technical specification. The export format must always be specified — these tools support different formats and the default is not always the most compatible.
  • Style keyword firstrealistic, stylized, low-poly, hard surface, organic, cartoon, photorealistic scan
  • Subject and key features — describe the primary object and its distinctive characteristics
  • Material and texturepolished chrome, worn leather, rough stone, iridescent scales, brushed aluminum
  • Technical spec — polygon budget (<10k tris for game-ready), texture resolution (4K PBR textures)
  • Negative prompt — exclude unwanted features: no floating parts, no disconnected geometry
  • Specify export formatGLB (web/Three.js), FBX (Unreal/Unity), STL (3D printing), OBJ (general)
Style: realistic, game-ready hard surface
Subject: medieval knight helmet, full-face visor with narrow vision slits, cheek guards with rivets, battle damage with dents and scratches on the left side
Material: aged steel, rust staining near the joints, dark patina on flat surfaces, polished highlights on raised edges
Technical: under 8,000 triangles, 4K PBR texture maps (albedo, normal, roughness, metalness)
Negative: no floating geometry, no disconnected parts, no organic forms
Export format: FBX

Export Format Reference

FormatBest For
GLBWeb (Three.js, Babylon.js), AR/VR
FBXUnreal Engine, Unity, Maya, 3ds Max
STL3D printing (geometry only, no texture)
OBJGeneral compatibility, Blender import
USDZApple AR Quick Look, iOS/macOS

Unity AI (Unity 6.2+)

Unity AI integrates directly into the Unity Editor through a command interface. Three slash commands drive the interaction: /ask for questions, /run for task execution, and /code for code generation.
  • /ask — use for questions about Unity APIs, best practices, or debugging guidance
  • /run — use for tasks that require Unity AI to perform editor actions (creating assets, configuring components)
  • /code — use for C# script generation; specify component type, behavior, and Unity version
/code
Generate a Unity 6.2 C# MonoBehaviour that handles player health.
Component: PlayerHealth
- int maxHealth = 100, int currentHealth starts at maxHealth
- Method: TakeDamage(int amount) — reduces currentHealth, clamps to 0, triggers OnDeath() if 0
- Method: Heal(int amount) — increases currentHealth, clamps to maxHealth
- UnityEvent OnDeath — assigned in Inspector
- Serialize all fields

BlenderGPT

BlenderGPT generates Python scripts that run in Blender’s scripting environment via the Blender API (bpy). Every prompt must specify geometry, material, and scene context — without these, the generated script will overwrite your scene defaults or produce geometry incompatible with your existing setup.
  • BlenderGPT generates Python scripts — the output is a bpy Python script, not visual instructions
  • Specify geometry contextcreate from scratch, modify existing mesh named "Cube", add to scene at origin
  • Specify material contextcreate a new Principled BSDF material, modify the existing material named "Metal_01"
  • Specify scene contextcurrent scene has one point light and a camera at (0, -10, 5), empty scene
  • Name objects explicitly — reference exact Blender object names to avoid modifying the wrong mesh
Generate a Blender Python (bpy) script for Blender 4.x.

Task: Create a procedural stone pillar mesh from scratch.
Geometry: Cylinder base, 8 sides, height 4m, radius 0.3m. Add a Loop Cut at the midpoint. Scale the top loop to 0.85x to create a slight taper.
Material: Create a new Principled BSDF material named "Stone_Pillar". Set Base Color to (0.35, 0.33, 0.30). Roughness 0.85. Specular 0.1.
Scene context: Empty scene. Place the pillar at world origin (0, 0, 0).
Output: Complete bpy script with comments. No UI operators — use bmesh or direct bpy.ops only.

Build docs developers (and LLMs) love