How to Build a Look Bible for an AI Film or Commercial

What goes in it

Keep it to one page. Each line should be something you can paste into a prompt or check with an eyedropper.

• Palette: Five to seven colors, each with a name, a hex code and where it lives in the frame, such as walls, wardrobe, sky or product.
• Light: Direction, hard or soft, color temperature in Kelvin, time of day, how dark the shadow side falls, and which practical lamps are allowed in shot.
• Lens and camera: One focal length per shot size (for example 35mm for wides, 85mm for faces, a 100mm macro for product), the aperture, camera height and the moves you allow.
• Texture: How much grain and which film stock it imitates, how sharp, whether highlights bloom.
• Format: Aspect ratio, frame rate and every delivery size.
• Grade: Contrast, black level and the name of the LUT file you will apply in the edit.
• World: Locations, era, wardrobe rules and props that recur.
• Never list: What the model adds when the bible is silent, such as lens flares, a teal and orange grade, or stray signage text.

For the format lines, use the standards your delivery will be judged on. Netflix's list of standard display aspect ratios runs from 1.78 (16:9) and 1.85 to 2.39 and 2.40, while social placements want 9:16 for Reels and TikTok and 4:5 for the Facebook feed. The Digital Cinema Initiatives specification requires support for 24 frames per second, and Google's Veo 3.1 generates at 24 fps, so 24 is the safe choice for a film that may screen in a cinema.

Write the lines the way a model reads them

The agency 100 Creatives kept count of what worked across its 2026 campaigns. The prompts behind kept images ran about 140 to 200 words on Google's Nano Banana models and about 300 on OpenAI's GPT Image. They opened with the subject and a pointer to the reference image, named the background with a hex color, described light as something falling on a surface, listed the palette as names plus hex codes, gave camera and lens one sentence, and put grain in a separate sentence of its own. That is a look bible, broken into prompt order.

Here is one of those kept prompts, for a food ad on white, made on Nano Banana 2 (lightly edited: brand names removed).

The product shown in the reference image sits perfectly centered on a white surface (#FFFFFF) surrounded by concentric rings of cocoa dust radiating outward like a bullseye target, created by a burst of powder frozen in perfect circular formation. Rich cocoa tones (#3E2015) grade outward to lighter cocoa dust brown (#B07D46) at the outermost ring. Sharp overhead flash renders every granule of cocoa powder in crystalline detail while casting a tight shadow directly beneath the brownie. The crumb texture of the brownie's top surface echoes the dusty rings below. Framed directly overhead at 50mm, f/8, focus even across the entire flat plane, composed as a perfect circle radiating from center. Heavy visible film grain throughout. Color grading references medium format Hasselblad digital, clinical sharpness, neutral whites, controlled warmth in the brown spectrum, pronounced grain, high contrast at center, saturation naturalistic and grounded in earth tones.

A person added the sentence "Heavy visible film grain throughout." by hand. All ten kept images on that job asked for heavy film grain, and nine carried that exact sentence. Only 14 of the job's 120 prompts carried it, so 9 of those 14 became keepers, against 1 of the other 106. Because it was added by hand, it marks prompts someone was already working on closely, so it shows the value of a locked line more than the power of four words. The same lesson holds for a film. Once a look line works, paste it unchanged into every shot.

Use style references with care

Reference images can carry a style faster than words, and they can carry too much. Midjourney, now on V8.2 by default, says its Style Reference "doesn't copy objects or people, just the overall style," with a style weight (--sw) from 0 to 1000 and a default of 100. Runway's References take up to three images per generation, and its guide suggests adding an image of the style you want alongside your prompt. Kling's IMAGE 3.0 takes up to 10 reference images and says it "can deeply deconstruct the core style of reference images." Google's Veo 3.1 reference images are for subjects, a person, character or product, and it lists no style type.

Use style references to find the look, then write down what they gave you. A style code or a reference image is hard to version and easy to lose, while a sentence of palette, light and grain goes into every prompt and every tool. If the reference is someone else's image, read our guide to using a mood board without copying it first.

Test the bible on a strip of stills

Before production, render six stills that cover the range of the film: a wide, a close-up, a product or prop insert, a night or low-light shot, the brightest shot and the darkest shot. Lay them side by side at the same size. If one looks like it came from a different film, the bible is missing a line; add it and render the strip again.

Then version it. Name the file with a number and a date, and write the version into your shot list, so a shot rendered on version 2 is not cut next to one from version 4. For a contest with a making-of or statement field, the bible also gives you a clear account of how you directed the look.

Carry it through the edit and grade

Generation gets the shots close, and the edit and grade make them match. ARRI describes a LUT as something that "transforms a digital signal to adjust the sensitivity and gamma, and provide a color correct monitoring input." In practice, one LUT or one grade applied to every shot pulls separate generations toward the same contrast and color. Grade after the cut is locked, check skin and product colors against the hex values in the bible, and add grain in the grade as well as in the prompt so every shot shares one texture. Our post-production guide covers the rest of the finishing chain.

What juries notice

Contest juries score the thing a look bible protects. The Runway AI Festival's rules score each criterion from 1 to 10, and its film track judges "Cohesion of the narrative & artistic message" while its advertising track judges "Cohesion of the concept & artistic message." Higgsfield's festival rules judge visual craft on "composition, cinematography, world-building, coherence of AI-generated imagery." Curious Refuge's 2026 Feel Good competition judged "uniqueness, consistency, and overall impact," and the Reply AI Film Festival scores craftsmanship as "overall quality of the production." A film whose shots look like they came from different films loses points on every one of those lines. Our guide to how to win the Runway AI Film Festival goes through one jury's criteria in detail.

How Overs writes a brand's look bible before it makes a photo

Overs works the way this page says to work. It writes the look down before any image exists. You give it one product photo and the brand's details, meaning its website, photos it already has and, if there is one, a brand guidelines PDF of up to 24 pages. It researches the brand and writes its visual DNA, the colors with hex codes, the light, lens, grain and styling, then runs art direction, casting and shot planning on top. Brand rules are applied last, so they have the final say, and you can watch every step.

The bible stays current. Brand Watch reads the brand's recent Instagram posts, on request at most once a month on the free plan, weekly on Pro and daily on Team, and every approval and rejection becomes brand memory for the next set.

The process comes from 100 Creatives, the agency Abhi Chawla founded, which used it to make hundreds of photos a week for hundreds of brands, and the brownie job above is one of its campaigns. Across four of them, 361 renders gave 45 keepers, about 8 tries per usable photo, which is $0.27 to $1.69 in AI fees per keeper at published prices of $0.034 to $0.211 a render. Abhi's own businesses make their images with Overs today. It makes stills, including key frames to animate, and writes a text motion prompt for each photo for Seedance or Veo. Overs' guide to a brand's visual DNA includes a template.

Overs is a sister product to AI Film Contests, from the same founder. The free plan covers 40 photos a month on your own OpenRouter key, with no markup from Overs. Try Overs free at www.overs.studio and read your brand's visual DNA before a single photo renders.