Reference-Based Color Grading for Ad Films And Commercials: Drop a Still, Match the Look

Reference-based color grading for ad films and commercials means dropping a still from a Park Pictures spot, a Greig Fraser brand film, or a photographer's print into your grading tool and matching your ARRI Alexa or RED ProRes footage to it — no LUT building required. For a 15-60 second commercial cycling through agency-client-director review, it collapses look-development from a full day to roughly thirty minutes.

I've been a colourist for four years and a DaVinci Resolve Certified one for most of that. I've graded ad films for Puma and WHSmith, plus a stack of branded content for India-based shops servicing global campaigns. The commercial space is its own animal: 15-60 second runtimes, three layers of approval, brand-mandated Pantones, and a 5-day turnaround that assumes nothing goes wrong. Reference-based grading is the only thing that keeps the look conversations from eating the schedule.

Why reference images are the only language the agency-client-director triangle actually speaks

Anyone who's sat through a commercial review knows the vocabulary problem. The agency creative pulls up a Park Pictures reel and says "this kind of energy." The director references a still from a Greig Fraser brand film — Apple, Vuitton, whatever campaign was current that month. The client, who isn't visual at all, says "more like that Coke thing from the Super Bowl." Three people, three references, and you're supposed to translate that into a grade by Friday.

If you're building primaries from scratch each round, you're negotiating an abstraction. "More teal" means one thing to you and another to a brand manager whose only frame of reference is the last commercial they happened to watch. Reference-based grading turns that conversation into a deliverable. Everyone agrees on the still. The still becomes the contract. The grade either matches it or it doesn't.

That's why every senior commercial colorist I know keeps a reference library — Erik Wilson's UK commercial cinematography, Ridley Scott Associates' signature work, stills from whatever campaign is winning at Cannes Lions that year. The library is the actual creative asset. The grading software is just the tool that resolves the reference into pixels.

What the reference match is actually doing under the hood

A good reference-match tool isn't doing magic. It's analyzing the reference image's color statistics — the distribution of values across highlights, midtones, and shadows; the dominant hue of the skin tones; the contrast curve; the saturation envelope per hue band. Then it computes a transform that pushes your source frames toward those same statistics.

The math is well-understood. The variable is how the tool handles edge cases: what happens when your reference is a still from a 35mm Kodak Vision3 print and your source is BRAW pushed through a custom CST? What happens when the reference is hard sunset light and your source is overcast 4pm? A weak tool will crush your blacks trying to match the reference's contrast. A good one gives you an intensity slider and lets you back off to 60% so the match becomes a starting point, not a finished grade.

The 5-day commercial schedule, honestly

The standard India commercial turnaround I work to looks like this:

• Day 1: Ingest, color space transform from Log to Rec.709 working space, scene cut into a shot timeline, drop the reference still, first-pass match across the entire spot.
• Day 2: Pull the rough cut into Resolve. Round-trip the pre-grade as a starting LUT or CDL. Refine primaries on hero shots — exposure, contrast, white balance. Identify brand-Pantone shots that need surgical secondaries.
• Day 3: Internal review with director. Adjust based on feedback. Send to agency.
• Day 4: Agency notes. Client notes via agency. Revise. Re-render proxies for review.
• Day 5: Final approval, deliverables. Broadcast spec, social cutdowns, behind-the-scenes grade.

Day 1 is where reference-based AI grading rewrites the calendar. Historically, day 1 was eight hours of look-dev — building nodes, balancing the hero shot, copying the grade to similar shots, fighting mismatched exposures across coverage. The agency saw nothing until day 2 evening. With a working reference-match pass, day 1 can produce a watchable, reference-matched cut by lunch. The director sees the look the same day. Course-corrections happen before you've sunk a full day of node work.

If you're an ad film colorist, we're building this for you. Leumos AI launches in ~30 days — join the early-access list and you'll be in the first 500 (50% off the first year).

Where reference matching breaks — and the workarounds

I won't pretend this is a one-click solution. Reference-based grading has predictable failure modes, and the colorists who win in commercial work are the ones who know where to stop trusting the AI:

Skin tones in mixed light. If your reference has clean key-lit skin and your source has a subject standing under mixed sodium-and-LED light, the match will skew the skin warm in a way that fights the existing color cast. You'll need to mask the skin and grade it separately in Resolve.

Brand Pantone compliance. Reference matching solves the look, not the legal. If the brand spec says Coca-Cola red must hit a specific Pantone value within tolerance, you have to qualify that color and push it manually. No AI tool reads brand bibles.

Day-for-night and other creative inversions. If the reference is a moonlit exterior and your source is a noon shoot — that's not a color grade, that's a VFX pre-comp. AI won't save you here.

Reference compression artifacts. If your reference is a 720p screengrab from YouTube, the color is already mangled. Garbage in, garbage out. Source references from BluRay rips, official campaign stills, or photographer portfolio sites at full resolution.

What I'm building, and how it fits the commercial workflow

I'm building Leumos AI to be the day-1 tool — the thing that takes a freshly delivered batch of ARRI Alexa or RED ProRes, transforms it to Rec.709, cuts it into shots, and matches it to a reference still before the director's first review.

The features that matter for ad film work specifically:

• Reference Image Grading — drop a still from a Greig Fraser spot, an Erik Wilson commercial, or a photographer's reference, and the AI pushes your footage toward it. Intensity slider so you can back off when the match gets too aggressive.
• Match All — once your hero shot is dialed, the AI equalizes exposure, contrast, brightness, saturation, and hue across every shot on the timeline. This is the multi-cam consistency problem in one click.
• AI Scene Cut Detection — uploads auto-chop into a shot timeline with thumbnails. No manual node-per-clip setup. For a 60-second spot with 40 cuts, this saves the first thirty minutes of every grade. The Manual Cut Tool catches the transitions the AI misses.
• Input Color Space LUT — S-Log3, C-Log3, BRAW, V-Log into Rec.709 in one click. The first thing you do on any commercial.
• Manual Primaries — when the AI gets you 80% there, you need surgical controls for the last 20%. Exposure, Contrast, White Balance (Temperature + Tint), and Saturation, per-shot.
• Preset LUT Library — curated cinema LUTs with intensity slider, plus support for your own .cube uploads when the brand has a house LUT.

Leumos won't replace Resolve for the finishing pass. Conform, deliverables, VFX integration, broadcast QC, secondaries for skin and brand color — that all still lives in the proper grading suite. Leumos is the front-end that makes day 1 productive instead of exploratory, so you arrive at day 2 with a real look to refine, not a blank timeline.

When Leumos launches in roughly 30 days, the first 500 early-access signups get 50% off the first year. If your week is structured around the 5-day commercial turnaround, that day-1 collapse is the difference between making the schedule and chasing it.

Frequently asked questions

What's the difference between reference-based color grading and using a LUT?

A LUT is a fixed transform — same input always produces the same output, regardless of what your footage actually looks like. Reference-based grading analyzes both your source and the reference image, then computes a custom transform that pushes one toward the other. For commercial work, that matters because every project has a different reference. A LUT pack can't anticipate the still your director will pull from a Park Pictures reel next Tuesday. Reference matching adapts to whatever the agency hands you. The two work together — a reference match gets you 80%, a LUT or your own primaries finishes the look.

Can AI reference grading handle Pantone-mandated brand colors?

Not directly, and you shouldn't trust any tool that claims it can. Reference matching works on overall image statistics — it'll get your spot to feel like the reference, but it won't legally guarantee that the Coca-Cola red hits a specific Pantone value within broadcast tolerance. That's a qualifier-and-secondary job, and it still belongs in Resolve. The workflow I run: AI reference match for the overall look on day 1, then on day 2 I open Resolve, qualify the brand color, and push it to spec manually. The AI saves the look-dev hours; the secondary work for compliance still takes a colorist.

Does reference matching work with ARRI LogC and RED Log3G10 sources?

Yes, provided you transform to a normalized working space first. ARRI LogC, RED Log3G10, Sony S-Log3, Canon C-Log3 — these are all log gammas. Reference matching on raw log footage produces unpredictable results because the statistics are non-linear. The right order is: transform your source from log to Rec.709 using an input color space LUT, then run the reference match against the normalized image. This is exactly why Leumos has a one-click Input Color Space LUT before the reference grading step. Skip the transform and the match will look muddy and contrast-crushed.

Will AI reference grading replace my Resolve workflow for commercial finishing?

No, and any tool claiming otherwise is overselling. Resolve still owns the finishing pass: conform from the offline edit, VFX round-trips, secondaries for skin and brand color, broadcast deliverables, QC, and DCP creation for cinema spots. Reference-based AI tools are front-end accelerators — they compress the look-dev phase that historically ate day 1. The realistic workflow is AI for the pre-grade and consistency pass, then Resolve for everything that has to be frame-accurate and broadcast-compliant. Pretending otherwise will get you in trouble the first time a commercial fails QC at the agency.

What resolution and quality should my reference stills be?

As high as you can source. A 720p YouTube screengrab has already been color-compressed twice — once by the upload encoder, once by the player. Reference matching against that gives you a match to the compression artifacts, not the original grade. Source references from BluRay frame grabs (1080p minimum, ideally 4K), official campaign stills from director or agency portfolios, photographer print scans, or full-resolution stills exported from Vimeo Pro. For brand work, the photography accompanying the commercial campaign is usually colorgraded to match — a single press photo from the campaign is often a better reference than a frame from the spot itself.

Can I match digital footage to a film-stock reference like Kodak Vision3?

Partially. Reference matching will pull your digital footage toward the film stock's color statistics — the warm midtones, rolled-off highlights, slightly desaturated shadows. What it can't replicate is grain structure, halation, and the non-linear color crosstalk that real film does. If the agency wants a true film look, the workflow is: reference-match for the base color, then layer a film emulation LUT or plugin on top for grain and halation. Tools like Dehancer or FilmConvert handle the texture side. The reference match handles the color side. Two passes, two tools, but the look ends up closer to actual film than either alone.

How does reference-based grading fit a 5-day commercial turnaround?

It restructures day 1. The traditional schedule loses eight hours to look-development — building nodes from scratch, balancing the hero shot, propagating the grade across coverage, fighting mismatched exposures. With AI reference matching plus an auto-equalization pass across the timeline, day 1 produces a watchable first cut by lunch. The director reviews the same afternoon instead of waiting until day 2 evening. That earlier feedback loop is what makes the agency review on day 4 less painful — you've already had two days of refinement instead of one. The 5-day window stops being a sprint and starts being a real schedule.

---

Leumos AI launches mid-2026. The first 500 early-access signups get 50% off the first year. Join the early-access list →