In professional filmmaking, lighting directors spend years learning to control light. It sets mood, defines the time of day, communicates emotion, and separates amateur-looking footage from cinematic masterpieces. In AI video generation, your prompt is the lighting director — and the words you choose determine everything.
This guide breaks down the most effective lighting descriptors for AI video prompts, with example prompts showing each technique in action.
Why Lighting Matters So Much in AI Video
AI video models like Runway ML and Pika Labs are trained on enormous libraries of film, photography, and video footage. These models have learned to associate specific linguistic descriptions with specific visual qualities. When you write "golden hour cinematography", the model activates a deeply embedded understanding of warm, directional, amber-toned light and the specific way shadows behave in late afternoon sun.
Poorly specified lighting leads to "averaged" output — flat, neutral illumination that lacks mood or direction. Precisely specified lighting produces scenes that look like they were shot by a skilled cinematographer.
The Essential Lighting Dictionary for AI Video
☀️ Golden Hour
The 30–60 minutes after sunrise and before sunset. Warm, amber, directional light with soft, long shadows. This is the most universally flattering light for outdoor scenes. AI models recognize this term expertly.
Keywords: golden hour, magic hour, warm directional sunlight, amber afternoon light
🌆 Blue Hour
The brief window of twilight after the sun drops below the horizon. Cool, soft, diffused blue-purple light that perfectly balances with artificial light sources. City scenes at blue hour look almost impossibly beautiful.
Keywords: blue hour, dusk, civil twilight, soft twilight, indigo sky
🌃 Neon Noir
Coloured artificial light sources — neon signs, LED strips, holographic displays — casting pink, cyan, and purple against dark urban environments. The defining aesthetic of cyberpunk and neo-noir genres.
Keywords: neon-lit, cyberpunk lighting, coloured artificial light, LED glow, holographic illumination
🎭 Chiaroscuro
High-contrast light and shadow, inspired by Caravaggio and Rembrandt. Large portions of the frame are in deep shadow, with a single harsh light source carving illumination across the subject. Extremely dramatic and theatrical.
Keywords: chiaroscuro, Rembrandt lighting, high contrast, single key light, deep shadows
🏔️ Overcast Diffused
The flat, even, colourless light of an overcast day. No harsh shadows, no bright highlights. Surprisingly one of the most emotionally versatile lighting conditions — can feel melancholic, peaceful, or ominous depending on context.
Keywords: overcast, flat natural light, diffused daylight, grey sky, soft ambient light
🕯️ Practical / Firelight
Light sourced entirely from practical elements visible in the scene — candles, fires, lamps, screens. Creates organic, flickering, warm illumination. AI models handle this with surprising accuracy when the source is specified clearly.
Keywords: candlelight, firelight, practical lighting, warm flickering glow, bonfire illumination
Combining Multiple Light Sources
The most sophisticated prompts layer multiple light sources to create complex, interesting illumination:
This prompt uses three light sources — spotlight, coloured wash, and practical smoke haze — creating dimensional, professional-looking illumination that simple prompts cannot achieve.
Lighting Phrases to Avoid
Some lighting descriptors are too vague to produce reliable results:
- "Good lighting" — Meaningless to the model.
- "Normal lighting" — Results in flat, uninspired output.
- "Dramatic lighting" (alone) — Too broad. Specify the type of drama.
- "Dark but visible" — A contradiction that confuses the model.
Always specify the light source, colour temperature, and direction wherever possible.
Practice With Our Library
Every prompt in our gallery includes carefully considered lighting descriptions. Study how different lighting terms are used within complete prompts to develop your own intuition for what works.