What cinematic color grading styles work best in AI video prompts — and how do you use them?

Specific, named grades outperform the word 'cinematic' every time: teal-and-orange for blockbuster contrast, split-toned amber and emerald for moody romance, low-key 85:15 dark-to-light grades for horror, warm Kodachrome-style fades for nostalgia. Write the grade as exact palette language in every prompt, then lock it as a fixed style block so every shot inherits it.

Drop 'cinematic' as a standalone descriptor — it collapses into generic stock-footage color. Replace it with three layers written in sequence: color temperature, the named palette, and tonal behavior (where the blacks sit, where the highlights roll off). One documented production went further and encoded a director's color philosophy as named tonal modes with exact hex values — 'Mode A — split-toned amber and emerald' — which made the grade reproducible across every generation of a 70-second short film finished in 2 days for $750.

Grades that reliably work in prompts, mapped to story tone:

• Teal-and-orange — 'warm amber skin tones against cool teal shadows, complementary palette.' The default blockbuster/action look; high subject-background separation.
• Split-toned amber and emerald — warm tungsten highlights, green-leaning shadows. Moody romance and neon-lit interiors; this is the mode-with-hex-values grade from the production above.
• Low-key thriller/horror — 'crushed blacks, 85:15 dark-to-light ratio, neutral gray shadows.' One horror production stored that exact 85:15 ratio as prompt language; when shadows drifted blue-green instead of neutral gray, the deviation was flagged against the locked rule and corrected with a warmer pass.
• Technicolor — 'vibrant saturated primaries, deep blacks.' Heightened, stylized period drama.
• Kodachrome-style warmth — 'golden highlights, rich greens, slight fade.' Nostalgia, memory sequences, indie drama with lifted shadows.
• High-contrast monochrome with selective color — graphic-novel and noir framing.
• Hand-painted desaturated painterly — for animation styles; one animated episode's style block read 'every surface has hand-painted brushstroke texture' with explicit prohibitions on live-action and photorealistic output, because negative constraints are what stop the model drifting back to default color.

How you place the grade matters as much as which one you pick. One documented protocol assembles every prompt in a fixed 9-element order — camera, lens and your film's aspect ratio, lighting source, palette, composition, atmosphere, mood, film attribution, negative prompt — so the palette always sits immediately after the lighting source. Tie color to its source rather than using floating adjectives: 'warm yellow from the lamps only, like all the refs' produces a measurably more accurate grade than 'warm lighting.'

For consistency across a whole film, write the grade once as a fixed style block and start every prompt with it. The team behind a 3-minute animated episode locked their style block after uploading 64 reference frames in a single message and opened all 164 clip prompts with it — that repetition, not any single prompt, is what held the palette across the entire episode. invideo is an agentic video creation tool with all the current video models available, and the invideo agent holds that color block in persistent context — you set the grade once and it carries into every shot's prompt automatically, whether the shot routes to Veo, Kling, or Seedance 2.0. The invideo agent's documented color grading guidance runs as an 8-step process, so you can also ask it to build the grade language for you from a description of the mood.

When you have a reference frame whose grade you want, don't paste a stylized or illustrated image directly as a generation reference — it transfers the illustration, not the color. Instead, instruct the invideo agent to read the palette and texture qualities of the frame and translate them into photorealistic prompt language; in one production the results came back hyper-realistic with the exact color temperature the director wanted. This reverse-engineering step is how you migrate an existing LUT or graded still into prompt form. If the generated footage still reads slightly plasticky, a light post pass — a touch of blur, grain, and a final grade adjustment — pulls it the rest of the way toward live action.

Watch some of these to see what works for you:

The better move was to have Agent 1 read the colours and textures of them and prompt for that instead.

— invideo's creative team, on translating reference frames into color-grade prompt language