Cinematic Lighting Prompts: Describe Light Like a Director, Not a Thesaurus

"Cinematic lighting" is the most common phrase in AI image prompts, and it does almost nothing.

The model reads it the way a stock photo site reads it: a bit of contrast, some teal and orange, maybe a lens flare. You get something that looks like a trailer thumbnail. Not like a frame from a film.

The fix isn't a better adjective. It's describing light the way a crew actually builds it on set. Where it comes from, which way it falls, how hard it is, and how much of the frame you let go dark.

Why "cinematic" fails

On a set nobody says "make it cinematic." The director of photography says something like: one hard source through the window, camera left, let the far side of the face fall off, keep a little bounce so the eyes don't die.

That sentence is specific about four things:

• Source — what is making the light (a window, a practical lamp, the sky, a street light).
• Direction — where it comes from relative to the subject and the camera.
• Quality — hard or soft, which depends on the size of the source.
• Contrast — how dark the shadow side is allowed to get.

Models were trained on millions of captions and behind-the-scenes descriptions that use exactly this vocabulary. When you give them the four variables, they stop guessing.

The four-variable lighting prompt

Here's the skeleton we use. It works in Midjourney, Nano Banana, Flux and as the lighting line of a video prompt:

[subject and action], lit by [source] from [direction], [hard/soft] light, [contrast: deep shadows / gentle falloff / high-key], [color temperature], [one atmospheric detail]

A real one:

A woman reading at a kitchen table at dawn, lit by a single window from camera left, soft light, deep falloff on the right side of her face, cool blue daylight against a warm tungsten practical in the background, faint dust in the air

Compare that with "a woman reading at a kitchen table, cinematic lighting, moody." The second one gets you an image. The first one gets you a decision.

Source: name the thing that makes the light

Light needs a reason to exist in the frame. When the prompt names the source, the model places it, and the shadows start to agree with each other.

Useful sources to name:

• A window — directional, usually soft, reads as natural.
• A practical — a lamp, a screen, a neon sign that's visible in the frame. Great for night scenes.
• Overcast sky — huge soft source, almost no shadows. Good for portraits that need to feel honest.
• Low sun — hard, warm, long shadows. Golden hour without saying golden hour.
• Car headlights, a fridge, a phone screen — motivated light for specific moods.

If you only change one thing in your prompts this week, make it this.

Direction: front light flattens, side light shapes

Front light is the default of every phone camera, and it's why so many AI portraits look like passport photos. Moving the source to the side is the fastest way to get shape.

• Side light (90°) — half the face lit, half in shadow. Dramatic.
• Three-quarter light (45°) — the classic portrait setup. Shape without drama.
• Backlight / rim light — the source behind the subject. Separates them from the background, outlines hair.
• Top light — harsh, interrogation room, midday sun.
• Under light — horror. Use on purpose or never.

Write it from the camera's point of view: "from camera left," "from behind the subject," "from above."

Quality: the size of the source decides it

Hard light comes from a small source: the sun, a bare bulb. It makes crisp shadow edges.

Soft light comes from a big source: a window with a curtain, an overcast sky, a bounce off a white wall. Shadows fade gradually.

Say it plainly: "hard light, crisp shadows" or "soft diffused light, gentle shadows." Then keep the source consistent with it. A bare bulb with soft light confuses the model the same way it would confuse a gaffer.

Contrast: decide how much darkness you can afford

This is the variable that most separates "film" from "photo." Films let things go black. Commercial photography fills every shadow.

• Low-key — most of the frame in shadow, one area lit. Thrillers, night interiors.
• Gentle falloff — light fades across the subject. Dramas, portraits.
• High-key — bright, minimal shadows. Comedy, beauty, product.

A phrase worth leaning on: "let the background fall into darkness." It gives the model permission not to light everything.

Color temperature: two colors beat one

A single color cast looks like a filter. Two temperatures in the same frame look like a location.

The classic combination is cool daylight from a window against a warm practical inside the room. At night, it flips: warm street light against cold moonlight or a blue screen.

"Teal and orange" is the overused version of this idea. Naming the actual sources gets you the same separation without the cliché.

The atmospheric detail

One small physical detail sells the light: dust in a window beam, haze, rain on glass, steam from a cup. It gives the light something to hit, which is how you see light in real life.

Use one. Three turns the frame into a fog machine demo.

Putting it together: three prompts

Night interior, drama

A man sitting alone in a parked car at night, lit only by the green glow of the dashboard from below and a passing street light from the right, hard light, deep shadows, the background lost in darkness, rain streaks on the windshield

Portrait, honest

Close portrait of an old fisherman on a dock, overcast sky as a huge soft source from above and slightly left, gentle shadows under the brow, muted cool palette, fine sea mist

Product, controlled

A matte black perfume bottle on wet slate, single hard backlight creating a thin rim along the glass edges, low-key, black background, a few water droplets catching the light

Each one names a source, a direction, a quality, a contrast level and one detail. None of them say "cinematic."

What to do with the ones that work

Lighting is the part of a prompt you'll reuse most. A good setup works across subjects: the dawn kitchen window works for a cook, a painter, a kid doing homework.

That's why the prompt should be saved together with the image it produced. Six months later, the text alone tells you nothing about whether the falloff was right. The result does. Keep the prompt, the model and the output together — Super Prompts does exactly that, and it's free to start.

If you want to go further with the camera side of the frame, read why AI video prompts look like stock footage next.