🎞️ 30 Cinematic Lighting & Grade Prompts for AI Video Models

We isolated one variable — the lighting clause, nothing else — across the same base shot on four models: Seedance 2.5, Veo 3.1, Kling 3.0, and MiniMax H3. Same subject, same action, same camera move, every time. Only the lighting/grade line changed between runs. That's the method behind every entry below: not "here's a vibe," but here's the exact clause, what it actually does to color and contrast, and which models take it literally versus which ones shrug and default to flat daylight.

Thirty clauses, five categories, ready to drop into your own subject-plus-action-plus-camera template. If you haven't locked that base structure yet, our Seedance 2.5 prompt library has the skeleton — this page is the lighting layer that goes on top of it.

By the numbers#

• 5600K is the standard daylight white-balance reference; 3200K is the standard tungsten reference — the two anchor points every "warm" or "cool" lighting clause below is actually describing
• Golden hour sits close to the tungsten end of that range as the sun drops, which is why "golden hour" and "warm tungsten interior" produce visually adjacent renders even though one's outdoor and one's indoor
• Two stacked lighting descriptors held up consistently in our tests; three is where reroll rates climbed on every model we ran
• Seedance 2.5 tops out at 720p (not the 4K some listings claim), but its beat-timeline structure was the only pipeline where a soft gradient (blue hour fading into warm practicals) held together as one continuous move across the whole clip instead of resetting

Category 1 — Motivated time-of-day light (1–6)#

Time-of-day clauses set color temperature and shadow angle in one move. These are the highest-leverage single words you can add.

1. Golden hour backlight — subject rim-lit, long shadows toward camera, warm 3200K-adjacent wash. Reliable on every model; the most-searched lighting term for a reason.

2. Blue hour exterior — pale blue ambient, streetlights just switching on, no direct sun. Pairs cleanly with practicals (Category 3) since the sky provides fill and the lamps provide key.

3. Hard midday overhead — short, harsh shadows, high contrast, unglamorous. Underused; good for realism/documentary work where golden hour would look too polished.

4. Soft overcast daylight — diffused, near-shadowless, even skin tones. This is the clause to reach for on talking-head and UGC-style shots — v1 ran six for six clean on Seedance and Veo with zero rerolls.

5. Pre-dawn cold light — thin grey-blue, minimal contrast, everything slightly underexposed. Reads as tense or lonely without any mood adjective doing the work.

6. Magic-hour transition, warm to blue — a timed shift within one shot, sun dropping below frame as ambient cools. Only worth the complexity on 15s+ generations; on short clips it never fully resolves before the take ends.

Category 2 — Lighting ratio & classic setups (7–12)#

These describe contrast and shadow shape, independent of color temperature. Combine one from here with one from Category 1.

7. Low-key, single hard source — deep shadows, narrow key, most of the frame near-black. Thriller/noir default; do not also add "golden hour" — the two fight over shadow softness.

8. High-key, minimal shadow — bright, even, near-shadowless across the whole frame. Sitcom/comedy/product-explainer default.

9. Rembrandt lighting — 45-degree key producing a small triangle of light on the far cheek. Reads as literal instruction on Veo 3.1 specifically; other models treat it more loosely as "dramatic portrait light."

10. Three-point setup named explicitly — "key light camera-right, fill camera-left, rim backlighting the hair." Longer to type, but this is the single clause that most reliably fixed our flat, directionless failures — spell out all three positions rather than saying "three-point lighting" as a buzzword.

11. Chiaroscuro, extreme contrast — near-total black with one lit plane. Failed more often than it worked on models without a strong reference image locking the subject — pair with a character reference or the shadows swallow identity details entirely.

12. Bounce-soft key, no visible source — one soft directional light, no harsh edge. The "safe default" for interview-style shots when you don't want a specific mood reading either way.

Category 3 — Practicals: light sources in frame (13–18)#

Practicals are visible sources — the model renders both the light and the object casting it. Every prompt guide we checked, including Google's own Veo 3.1 guidance, flags these as reliably interpreted.

13. Warm practicals, cozy — lamps, candles, string lights visible in shot, amber cast on nearby surfaces. Google's own Veo guidance names this exact framing — warm sources read as cozy, cool sources read as clinical.

14. Cool practicals, clinical — monitor glow, fluorescent tube, blue-white cast. Office, hospital, sci-fi interior default.

15. Neon signage, mixed color — magenta/cyan practicals reflecting on wet or glossy surfaces. Needs a reflective surface named in the same prompt (wet asphalt, glass, chrome) or the color cast has nothing to bounce off.

16. Single desk lamp, hard-edged pool — one visible source, everything outside its throw falls to near-black. Cheap-feeling on purpose; good for intimate or covert scenes.

17. Car headlights sweeping through frame — moving practical source, brief hard backlight as it passes. Only worth it on models with real camera-timeline support (Seedance 2.5's beat structure); on single-beat generations it reads as one static flare instead of a sweep.

18. Firelight flicker — orange, unsteady, low-angle uplight on faces. Distinct from "warm practicals" because the flicker/instability is the point — say "flicker" explicitly or you get a static warm glow instead.

Category 4 — Film-stock emulation (19–24)#

Stock language describes how light lands on the image, not the light itself — always pair with a source clause from Category 1 or 3.

19. Kodak Vision3 500T tungsten stock — teal-leaning shadows, warm-orange highlights, visible grain. The most reliable stock reference across all four models we tested; name it alongside an interior practical for the intended effect.

20. Kodak Portra warm tones — soft contrast, warm skin, gentle highlight roll-off. Portrait/lifestyle default; underwhelming on anything without a visible face in frame.

21. Fujifilm Eterna, desaturated teal-green — muted palette, restrained color. Reaches for a specific "prestige drama" look; pair with soft overcast (Category 1) rather than golden hour, which fights the desaturation.

22. CineStill 800T halation — visible red-fringe glow around bright practicals at night, tungsten-balanced. Only shows up correctly when a bright point-source practical (Category 3, entries 15–17) is also in the prompt — with no bright source to halo, this clause does nothing visible.

23. High-contrast black-and-white, heavy grain — classic noir stock feel. Strip color-temperature words entirely once you add this — "golden hour black and white" produced confused, partially-desaturated results on every model in our test batch.

24. Vintage Kodachrome saturation — punchy primaries, slightly crushed shadows. Good for retro/nostalgia framing; reads as dated on purpose, don't combine with modern practicals like LED strips.

Category 5 — Grade closers (25–30)#

Close the prompt with one of these as the final line — the global style instruction the whole shot resolves to.

25. Teal-and-orange blockbuster grade — skin pushed warm, shadows/backgrounds pushed cool-teal. The most over-requested grade in our queue; works, but every model defaults toward it slightly even unprompted, so it's the least distinctive choice on this list.

26. Bleach-bypass, desaturated and gritty — retained silver, crushed color, high contrast. War/thriller default; needs hard or overcast light upstream, not golden hour — the softness fights the grade's harshness.

27. Clean commercial, high-key neutral — minimal color grading, accurate whites, no stylized cast. The correct closer for product/UGC work when the client wants "real," not "cinematic."

28. Day-for-night, cool blue crush — daylight footage graded to read as night, slight underexposure. Historically a film trick; on AI video it's actually easier to just prompt real night lighting directly — we kept this in the library because it still reads distinctly, but it's the one entry we'd skip for a first attempt.

29. Cross-processed, color-shifted — unnatural color shift, blown highlights, alt-reality feel. Music-video and dream-sequence use; unpredictable enough that we ran it in triplicate and got three visibly different grades from the same prompt on Kling specifically.

30. Documentary handheld, minimal grade — near-neutral, slight desaturation, no stylization. Pairs with the consistent-character workflow when the goal is believability over polish — this is the grade line we reach for most on B-roll.

When to use which#

• Product ads / UGC: soft overcast or clean commercial grade (4, 27) — anything moodier reads as trying too hard for a 15-second spot
• Documentary / interview: soft overcast, bounce-soft key, documentary handheld (4, 12, 30) — skip stocks entirely, they fight the "real" read
• Thriller / noir: low-key single source, hard-edged pool, bleach bypass or black-and-white (7, 16, 23 or 26)
• Romance / nostalgia: golden hour, warm practicals, Portra or Kodachrome (1, 13, 20 or 24)
• Sci-fi / clinical: cool practicals, chiaroscuro if a face is locked by reference, teal-and-orange as the safe fallback (14, 11, 25)

Failure patterns we cut#

Same rule that runs through every library we publish: specific beats buzzwords, every time. "Cinematic," "moody," and "epic lighting" alone underspecify the shot and every model we tested defaulted toward the same generic push-in look regardless of which one we typed. Stacking three or more lighting clauses in one prompt raised our reroll rate on every model — pick one time-of-day/source and one contrast clause, then close with a single grade line. And matching light to logic matters: a hard midday sun clause paired with heavy practicals confuses the render because the two imply different scenes, not one. If you're new to this stack generally, our beginner mistakes roundup covers the non-lighting version of the same problem — vague, stacked, contradictory instructions losing to short, specific ones.

Model notes#

None of this vocabulary is model-exclusive, but the read quality varies. Seedance 2.5 caps out at 720p, not the 4K some listings imply, but its beat-timeline structure keeps a subtle gradient (blue hour fading into practicals) reading as one continuous move instead of resetting mid-clip — worth the extra setup if you're doing a timed lighting shift. Veo 3.1 was the most literal with named techniques in our tests, treating "Rembrandt" or "rim light" almost like keyword triggers rather than loose suggestions. Kling 3.0 wants the lighting/mood clause as one of its four required prompt elements (subject, camera, light, setting) or it drifts toward a generic default — see our Kling motion-control library for how that structure plays out on a full prompt. If you're choosing a model for a lighting-heavy job specifically, Seedance 2.5 vs Veo 3.1 and Kling 3.0 vs Seedance 2.5 both break down where each one actually wins on render fidelity.

Log your own batch before you trust any of this on a paid render — a fixed base shot, one lighting variable changed per run, and a plain note on what failed. That's the only way a prompt library earns the word "tested" instead of "guessed."