AI Video Lighting Prompts: Cinematic Light Setups That Actually Work

You wrote the perfect subject prompt. The camera move nailed. The character is exactly who you imagined. And then it looks flat. Like a webcam recording. Like a TV commercial for cough medicine. Like anything but a movie.

The difference between “AI video” and “this looks like it came out of a real production” is almost always lighting. Cinematographers spend more time lighting a scene than blocking or shooting it. AI video models reward the same investment in your prompt — a specific, intentional lighting description shifts an output from generic to cinematic more than any other variable.

This is the lighting prompt cheatsheet for Sora 2, Veo 3.1, Kling, and Runway Gen-4 in 2026. The lighting setups professional cinematographers actually use, the exact phrases each model responds to, and copy-paste templates for the looks that matter most.

Why Lighting Matters More Than Camera in AI Video

A great camera move shot in flat fluorescent lighting still looks like security footage. A static shot lit with motivated practicals from a single window looks like the opening of a Wes Anderson film. The lighting carries 70% of the cinematic feel.

AI video models in 2026 have gotten dramatically better at interpreting lighting language. Sora 2 understands kelvin temperatures. Veo 3.1 was trained heavily on cinematographer terminology and renders motivated lighting accurately. Kling responds well to mood-based lighting descriptors. Runway Gen-4 handles dramatic contrast better than any other model.

The constraint is no longer the model — it is whether you know what to ask for.

The Four Lighting Variables to Specify in Every Prompt

For any lighting prompt, include all four of these:

1. Direction — where the key light comes from (left, right, above, behind)
2. Temperature — the color of light in kelvins or descriptive terms (3200K tungsten warm, 5600K daylight, 10000K cool blue)
3. Quality — hard or soft (a single spotlight vs a softbox vs natural diffusion)
4. Source — what’s making the light (window, lamp, sun, neon sign, candle, screen glow)

Specifying all four kills the model’s tendency to default to flat ambient lighting. Specifying only one or two leaves too much to the model’s guess.

Compare:

“A woman sits at a desk, bright lighting.”

Versus:

“A woman sits at a desk. Warm 3200K tungsten light from a desk lamp camera-right, soft fill from a window camera-left. The lamp throws hard shadows across her face while the window light wraps softly around the opposite side.”

Same scene. Same subject. The second prompt produces a controlled, intentional look. The first produces a generic stock-footage feel.

Setup 1: Three-Point Lighting (The Classic)

Three-point lighting is the foundation of every interview, dialogue scene, and product shot in commercial production. Key light, fill light, back light. Sometimes called the universal lighting setup.

Copy-paste template:

Subject: [your subject]. Lighting: classic three-point setup — a soft key light at 5600K from camera-left at 45 degrees, a softer fill light at 4800K from camera-right at 30% intensity, and a hair light from behind the subject above the shoulder line. Soft falloff on the background. Natural skin tones, no harsh shadows.

Use for: talking-head shots, product reveals, interviews, anything where you want the subject clearly visible without drama.

Model notes:

• Sora 2: responds well to “three-point” by name
• Veo 3.1: prefers explicit positions (“camera-left”, “camera-right”, “above”)
• Kling: prefers softer descriptors (“warm key”, “gentle fill”) over technical
• Runway: needs each light’s purpose stated (“for the face”, “for the hair”)

Setup 2: Golden Hour (The Always-Wins)

The hour after sunrise and before sunset. Low-angle sun, warm 2500-3500K color temperature, long soft shadows. Every cinematographer’s favorite light because it’s flattering and dramatic at the same time.

Copy-paste template:

Subject: [your subject] standing/walking in [your setting]. Lighting: golden hour, 30 minutes before sunset. Warm 2800K sun low on the horizon, hitting the subject from behind and slightly camera-right, creating a rim of light around their silhouette. Soft warm bounce light fills the shadow side. Long shadows stretch across the ground. Warm orange tone throughout, slight haze in the air catching the light.

Use for: outdoor scenes, romance, nostalgic feels, anything that wants emotional warmth.

The “haze in the air catching the light” is doing a lot of work in this prompt — it forces the model to render visible light beams (volumetric lighting), which makes the shot feel cinematic instead of digital.

Setup 3: Window Light (The Vermeer)

A single window as the key light source, soft and directional. Named after the Dutch painter for how its quality looks. Used in countless prestige dramas (Joanna Hogg, Yorgos Lanthimos, every A24 film).

Copy-paste template:

Subject: [your subject] sitting indoors. Lighting: single large window camera-left, soft overcast daylight at 6500K streaming in. The window light wraps around the subject’s face, falling off into deep shadow on the camera-right side. No fill light, no other sources. Walls and background fall into soft darkness. Dust motes visible in the window beam.

Use for: introspective moments, dialogue, portraiture, anything you want to feel still and contemplative.

This setup teaches you how much “lack of light” matters. Most AI video defaults to evenly lit. Specifying “no fill” and “background falls into shadow” creates the negative space that makes the lit subject pop.

Setup 4: Neon and Practical (The Cyberpunk)

Lighting motivated by signs, screens, and light sources visible in frame. The Blade Runner look, but also the look of any modern night-exterior city scene.

Copy-paste template:

Subject: [your subject] standing on a wet city street at night. Lighting: motivated by neon signs visible in frame. Magenta neon “OPEN” sign camera-left bathes the subject’s face in pink. Cyan storefront sign camera-right rim-lights the opposite side. Overhead streetlight at 2200K throws a small warm pool on the wet asphalt. Reflections in puddles double the color intensity. Slight fog in the air.

Use for: night exteriors, urban scenes, action sequences, anything you want to feel modern and saturated.

Model notes:

• Veo 3.1 handles colored lighting interactions the best — multiple colored sources mixing
• Sora 2 can struggle with neon color accuracy unless you specify the exact hue (magenta, not pink; cyan, not blue)
• Kling tends to dampen neon saturation; counter with “extremely saturated neon” or “vivid neon colors”
• Runway does dramatic neon contrast well but loses the wet-street reflections

Setup 5: Candlelight and Firelight (The Barry Lyndon)

Single-source low-color-temperature lighting. Named after Kubrick’s film famously shot by candlelight. Warm, intimate, naturally flickering.

Copy-paste template:

Subject: [your subject] sitting at a wooden table. Lighting: single candle flame in frame as the only light source. Warm 1800K flickering candle light hits the subject’s face from camera-right, creating dramatic shadow falloff to deep darkness on the opposite side. Soft warm bounce off the wooden table provides minimal fill. Background nearly black. The flame’s subtle flicker animates the shadows on the wall behind the subject.

Use for: intimate dialogue, period pieces, dramatic moments, anything with emotional weight.

The “subtle flicker animates the shadows” instruction matters because most AI video defaults to static lighting. Asking for the lighting to move with the source produces dramatically more believable output.

Setup 6: Overcast Diffused (The Documentary)

Soft, even, directionless light. The look of a documentary, a news interview, an outdoor scene on a cloudy day. Flat in the technical sense but never boring — it lets the subject and color carry the shot.

Copy-paste template:

Subject: [your subject] outdoors in a [setting]. Lighting: overcast afternoon, 6500K cool daylight diffused evenly through cloud cover. No hard shadows, no directional key. Subject is lit with soft wraparound light. Colors slightly desaturated. Slight cool tint to the shadows. Documentary realism.

Use for: documentary feel, realistic outdoor scenes, anything you want to feel “captured” rather than “produced”.

Setup 7: Single-Source Hard Light (The Noir)

One hard light source. Deep shadows. High contrast. The film noir look, also used heavily in modern psychological thrillers.

Copy-paste template:

Subject: [your subject] standing in a [setting]. Lighting: single hard 3200K light source from above and camera-right at a 60-degree angle. Sharp-edged shadows fall across the subject’s face — one side fully lit, the other side in deep shadow. The shadow of [an object/the subject] falls dramatically across the background wall. No fill light. Background is dark gray to black.

Use for: tension, mystery, antagonist reveals, anything dramatic.

This is the hardest setup to get right in AI video. Models tend to soften light automatically. Counter with explicit terms: “hard-edged shadows”, “no fill”, “harsh contrast”, “single source only”.

Setup 8: Volumetric Light (The God Ray)

Light beams visible in air thick with dust, fog, or atmosphere. Forces depth. Adds drama to nearly any setting.

Copy-paste template:

Subject: [your subject] inside [a large space]. Lighting: shafts of light cutting through the air at sharp angles. Volumetric god rays from [a window / a skylight / a doorway / above]. Dust particles visible in the beams. Where light hits surfaces, it pools brightly; where it does not reach, deep shadow. The visible light beams are the dominant visual element of the shot.

Use for: cathedrals, warehouses, forests at dawn, sci-fi interiors, anything that benefits from visible atmosphere.

The keyword “volumetric” is the magic word in 2026 AI video prompting. Including it forces the model to render light passing through atmosphere instead of just lighting the surfaces. The keyword pairs well with our AI video prompt structure formula for repeatable results.

Setup 9: Screen Glow (The Modern)

A subject lit primarily by the screen they’re looking at. Phone, laptop, TV. The defining lighting of the 21st century.

Copy-paste template:

Subject: [your subject] looking at a [screen type]. Lighting: subject’s face is lit only by the cool 6500K blue-white glow of the [screen]. The light is soft and comes from below the face line, creating subtle uplight shadows. Background is dark, almost black. Slight color variation as the screen content changes. Reflection of the screen visible in the subject’s eyes.

Use for: contemporary scenes, late-night moments, sci-fi, anything that wants to feel current.

Setup 10: Mixed Color Temperature (The Realistic Interior)

Real interiors are lit by multiple sources at different color temperatures — a 2700K incandescent lamp, a 4000K kitchen overhead, a 6500K window. Specifying this mix produces dramatically more realistic interiors than asking for a single source.

Copy-paste template:

Subject: [your subject] in [interior space]. Lighting: mixed practical sources. Warm 2700K table lamp camera-right pools yellow on one side of the room. Cool 6500K daylight from a window camera-left wraps the other side in soft blue. Where they meet on the subject’s face, the colors blend naturally. Overhead fixture is off. Believable real-world lighting mix.

Use for: any interior scene that should feel like a real place, not a soundstage.

Lighting Modifiers That Work in Every Setup

Beyond the basic four variables, these modifier terms shift the quality of any lighting setup:

• “Motivated lighting” — light that comes from visible sources in the scene (vs studio lighting added for effect)
• “Naturalistic” — realistic light behavior, soft falloff, accurate color
• “Stylized” — the opposite, push contrast/saturation/color in unrealistic directions
• “High key” — bright, evenly lit, minimal shadows (sitcom, commercial)
• “Low key” — dark, deep shadows, high contrast (noir, drama)
• “Chiaroscuro” — extreme high-contrast painting reference, very dark + very light
• “Practical lighting” — lamps and sources visible in frame
• “Bounce light” — light reflected off a surface back at the subject
• “Spill light” — the slight light reaching areas not directly lit
• “Rim light” / “backlight” — light from behind the subject outlining their silhouette
• “Top light” — harsh light from directly above (creates eye-socket shadows, dramatic)

Mix two or three modifiers per prompt for fine control without overloading the model.

Common Lighting Prompt Mistakes

Mistake 1: Asking for “good lighting” or “professional lighting”. Vague descriptors get vague results. Be specific about every variable.

Mistake 2: Defaulting to “bright” or “well-lit”. Most cinematic lighting is selective — dark in places, bright in others. “Well-lit” produces flat output.

Mistake 3: Specifying only the key light. Real lighting has fill, bounce, rim, and ambient. Even if you want a single dramatic source, name what’s not there (“no fill light”, “deep shadows in the unlit areas”).

Mistake 4: Mixing too many sources. Three to four light sources in a single prompt is the upper limit. Beyond that, the model averages everything into mush.

Mistake 5: Ignoring color temperature. “Warm light” is vague; “2800K tungsten” is specific. The kelvin specification is the single highest-leverage detail you can add.

Putting It Together: A Complete Prompt

Here is what a fully lit prompt looks like, combining everything above:

A young chef plates a dish in a darkened restaurant kitchen at the end of service. Medium shot, low angle, camera at counter height.

Lighting: motivated practical lighting only. A single overhead 3200K pendant lamp directly above the plating station throws hard light down onto the food and the chef’s hands. The chef’s face is in soft shadow, lit only by warm bounce off the white plate. Background kitchen is dark except for the distant green glow of an exit sign and the soft yellow spill from another lamp 20 feet away. Volumetric haze from cooking earlier in the night catches the overhead beam.

Naturalistic, low key, chiaroscuro mood. Color temperature mix between warm tungsten and the cool green exit sign creates depth.

That prompt produces output that looks like a real restaurant scene shot by a competent DP. Without the lighting detail, the same subject and camera produce something that looks like an AI demo.

If you are building a library of these for repeated use, LzyPrompt’s prompt generator saves and adapts lighting templates across Sora, Veo, Kling, and Runway — one input becomes four model-tuned outputs.

FAQ

Do I need to specify kelvin temperatures, or are descriptive terms enough? Both work, but kelvin temperatures are more precise. “Warm light” can mean anything from 2200K to 4500K depending on the model’s interpretation. “3200K tungsten” eliminates ambiguity. Use kelvins for technical control, descriptive terms when you want the model to interpret loosely.

Which model handles lighting best in 2026? Veo 3.1 currently has the most accurate cinematic lighting interpretation, especially for motivated practicals and color temperature mixing. Sora 2 is close, with slightly better volumetric lighting. Runway handles high-contrast and dramatic looks well but struggles with subtle naturalistic lighting. Kling is best at soft, flattering portrait-style lighting.

Can I prompt for specific cinematographers’ lighting styles? Yes, with mixed results. “Lit in the style of Roger Deakins” or “Hoyte van Hoytema lighting” gives the model a reference but the results vary by model. More reliable: describe the actual qualities of that DP’s lighting (Deakins = naturalistic, motivated, often single-source; Hoytema = practical-heavy, color-temperature mixed, atmospheric).

Why does my AI video sometimes ignore the lighting instructions? Usually because they conflict with stronger signals elsewhere in the prompt. Telling the model “outdoor sunny day” plus “dim candlelit interior” creates contradictions. Make sure your subject, environment, and lighting align. If the model ignores lighting consistently, reduce subject/camera complexity to give the lighting prompt more weight.

How long should a lighting prompt be? 20-60 words for the lighting portion of a full prompt is the sweet spot. Below 20 you have not specified enough. Above 60 the model averages everything together and you lose precision. If you need more detail, prioritize: direction, temperature, quality, source, then everything else.