AI Color Grading for Ad Films: A Colourist's Guide | Leumos AI

I've spent the last four years grading commercial spots — Puma, WHSmith, a stack of brand films for Indian and global clients you've probably seen during an IPL ad break. Most of those jobs landed in my Resolve project at 4pm on a Monday with a Thursday delivery, a Pantone deck from the brand team, an agency creative who'd already mocked up frames in Photoshop, and a director who wanted it to feel like Greig Fraser's Apple work but warmer.

If you've graded a commercial in the last decade you know the rhythm. Fifteen to sixty seconds of finished spot. Thirty to two hundred shots in the can. ARRI Alexa or RED footage transcoded to ProRes 4444 by the DIT. Three review rounds — director, agency, client — each with their own notes, half of which contradict the other half. And somewhere in there, the actual creative grade.

The part that's eating our lunch isn't the creative grade. It's the seventy percent of the day we spend equalizing footage before we can even start. That's the part AI color grading is genuinely good at now, and that's the part I want to talk about — not as a believer, but as someone who's burned a lot of weekends on this exact problem.

What AI color grading actually does on a commercial timeline

Let me be specific about what I mean, because "AI color grading" has become a buzzword salad. I'm not talking about one-click presets. I'm talking about three concrete things:

Shot-to-shot matching. You shot a beauty pack-shot at 5pm on day one and the hero close-up at 11am on day two. Same product, same lighting setup, but the ambient bounced differently and your DP changed the ND between setups. Equalizing those two shots used to be ten minutes of eyeballing scopes and dragging wheels. AI matching gets you 85% of the way there in a few seconds.

Reference matching. Your director sends you a frame from a Bradford Young commercial — the Jay-Z spot, or one of his AT&T films — and says "this, but for sneakers." Building a node tree to chase that palette manually is the bulk of a morning. Reference matching collapses it to a starting point in seconds. Not a finished grade. A starting point.

Log normalization. S-Log3, C-Log3, RED IPP2, BRAW. We all have our preferred CSTs and we all spend ten minutes setting up the project before we touch a single shot. AI tools that handle input transforms automatically save us from the most boring twenty minutes of every job.

What AI doesn't do — and won't do well any time soon — is the creative call. It can't decide that the brand wants warmer mids than the agency thinks they want. It can't read a room during a review. It can't tell you that the client's brand red is Pantone 186 C and your AI-matched grade is currently sitting at something closer to 200 C, which will get the spot kicked back.

Where the time actually goes on a 30-second spot

Let me break down where my hours go on a typical five-day grading turnaround for a brand film. I tracked this on the last twelve jobs.

• Day 1 morning: Project setup, CST configuration, LUT loading, basic conform — 2 hours
• Day 1 afternoon: Shot-by-shot equalization across the timeline — 4 to 5 hours
• Day 2 morning: Reference building, mood frame discussion with director — 2 hours
• Day 2 afternoon: Creative grade pass one — 4 hours
• Day 3: Director review + revisions — full day
• Day 4: Agency review + revisions, secondaries, skin tone work — full day
• Day 5: Client review, Pantone compliance pass, final QC, deliverables — full day

Day one afternoon is the obvious target. Five hours of dragging wheels to make shot 47 match shot 46. That's the part AI cuts down to under an hour. Not zero — you still need to check the AI's work and override it when it gets confused by motivated lighting changes or a wardrobe color that's tricking the algorithm. But under an hour, easily.

That means by Tuesday lunch you're already into the creative grade instead of starting it Wednesday morning. On a five-day job that's a full extra day of creative iteration. Which, in commercials, is the difference between a grade that wins the room and a grade that just ships.

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).

The Pantone problem (and why AI alone won't solve it)

Here's where I have to be honest with you, because this is the part that bites every colorist who tries to use AI on a brand job without thinking.

The brand wants the bottle to be Pantone 186 C. The agency has a brand guide with hex values, CMYK breakdowns, and reference renders. The client's marketing director will hold up a printed sample next to the monitor in the review and they will know if it's off.

AI color matching does not understand Pantone. AI scene matching also doesn't understand that the product is the only thing in the frame that's allowed to be that exact red — your model's lipstick can drift, the brick wall in the background can drift, but the bottle cannot.

What I do on every brand job: AI handles the global equalization pass and the rough creative match. Then I isolate the product with a tracked qualifier or a window, pull a HSL key on the brand color, and lock it manually. The AI gets me to a great starting point across the whole timeline in twenty minutes; the manual product isolation work still takes me forty-five. But that forty-five used to be on top of five hours of equalization, not on top of twenty minutes.

This is the workflow that actually works for commercial colorists right now. AI for global pass, manual for brand-critical elements. Anyone telling you AI does the whole job hasn't graded a real brand spot recently.

How I'd actually set up an AI-augmented commercial grade

This is the pipeline I've landed on after about a hundred test grades comparing different tools. It's tool-agnostic — same logic works whether you're using Colourlab, Leumos, or fylm.ai. The point is the order of operations.

Step one: input transform. Get everything to a working color space first. If you're on ARRI LogC3 going to Rec.709, do that conversion before you do anything else. Input Color Space LUT handles S-Log3, C-Log3, BRAW, V-Log in one click — pick yours and move on.

Step two: scene detection. Don't manually mark every cut. Upload the conformed timeline and let AI Scene Cut Detection chop it into a shot timeline. For the cuts it misses — usually fast in-camera whip transitions — use the Manual Cut Tool to fix them in seconds.

Step three: global match. Match All auto-equalizes exposure, contrast, saturation, and hue across the whole timeline. Don't expect perfection. Expect 80%. Spot-check the shots that look weird and override them.

Step four: reference grade. Drop your director's reference frame into Reference Image Grading. I've used stills from Roger Deakins' Skyfall night-exterior work as a starting point for a moody automotive spot — it gets you to the neighborhood. Dial the intensity slider down to maybe 60% so the AI doesn't crush your blacks the way it wants to.

Step five: manual primaries and secondaries. This is where you actually earn your day rate. Manual Primaries for exposure, contrast, white balance, and saturation adjustments. Then Resolve for the qualifier work, the product isolation, the skin tone surgery. Browser tools aren't there yet for tracked windows and complex secondaries — and they probably shouldn't be.

What about Colourlab, fylm.ai, and color.io?

They're all credible tools. Colourlab's matching model is genuinely impressive, especially on skin tones — they have a team that's been at this for years. fylm.ai's film emulation library is the best one I've used in a browser. color.io has a slick interface and good collaboration features.

What I've wanted, and what I'm building toward with Leumos, is something I can open in a browser at a client's office on whatever Mac they hand me, work fast on a 30-second spot without rendering proxies for ten minutes, and use as a first-pass tool before I round-trip into Resolve for finishing. Not a Resolve replacement. A pre-Resolve speed layer.

That's it. That's the whole pitch. If your work is mostly long-form or feature, Resolve plus a tool like Colourlab is probably still your stack. If you're doing 15-60 second spots on a tight turnaround and your bottleneck is the equalization pass, that's the gap we're filling.

Frequently asked questions

Can AI color grading actually handle the Pantone-compliance requirements on a brand spot?

Not on its own, and anyone claiming otherwise hasn't graded a real brand job. AI is excellent at global matching and getting your timeline to a consistent starting point. It cannot reliably hit a specific Pantone value on a product because it doesn't understand which element in the frame is brand-critical. The working pipeline is: AI for the global equalization and creative match, then a manual HSL qualifier or tracked window on the brand color with hand-tuned values. You'll still spend forty-five minutes locking the product color — but you save five hours upstream.

I shoot mostly ARRI Alexa ProRes 4444 — will AI tools handle that natively?

Yes. Alexa ProRes 4444 with LogC3 is one of the most common combinations these tools are trained on, so matching and reference work tend to perform best on this format. The first step is always the input transform — get LogC3 to Rec.709 with a proper CST or input LUT before you do any matching. RED IPP2 and BRAW are also well-supported now. Where AI tools still struggle a bit is heavily underexposed Alexa footage where the noise floor is messing with the matching algorithm — in those cases override the AI on individual shots.

How does this fit into a typical 5-day commercial grading turnaround?

The five-day rhythm doesn't change — director review on day three, agency on day four, client on day five — but you get back roughly a full day of creative time. The equalization pass that used to consume Monday afternoon collapses to about an hour Monday morning. That means you're into the creative grade by Monday lunch instead of Wednesday. On a 30-second spot with thirty to sixty shots, that extra day usually shows up as more iterations during the director review on day three, which is when grades actually win or lose.

Does AI matching work on mixed lighting situations — like a spot that cuts between tungsten interior and daylight exterior?

Partially. The AI will neutralize both shots toward a consistent white balance, which is exactly what you don't want if your creative intent is for the interior to feel warm and the exterior to feel cool. The trick is to do the AI matching pass first to get everything technically consistent, then go back and manually push the warmth back into the interior shots. Or, group your shots by lighting condition before running the match, so the AI is only matching tungsten-to-tungsten and daylight-to-daylight. Mixed-light skin tones are still the hardest problem in this space.

Can I use this for the final delivery or is it just a first-pass tool?

Honestly, for ad film work I'd treat browser-based AI tools as a first-pass and pre-Resolve speed layer, not a final delivery tool. The reason: secondaries, tracked windows, complex power windows, beauty work on talent — Resolve is still the industry standard for that, and your colorist supervisor or DI house expects to receive an XML or AAF they can round-trip. For social-cut deliverables and pitch grades where you're just trying to show the director a direction quickly, finishing in the browser is fine. For broadcast or theatrical deliverables, finish in Resolve.

What's the realistic time savings on a 30-second spot with 50 shots?

On my last three jobs at roughly that scale, I clocked the equalization pass at fifty to seventy minutes total instead of four to five hours. Reference grade setup dropped from about ninety minutes of node-building to fifteen minutes of dragging a reference frame and dialing intensity. Total saved time across the project was roughly six to seven hours, or basically a full day. The creative work, the review rounds, and the brand-color compliance work didn't change — those still take what they take.

I'm a senior colorist with a finishing room and a panel — is this actually for me?

It's for the part of your day you don't enjoy. If you love the equalization grind, no. But every senior colorist I know would rather spend Monday afternoon on creative looks than dragging wheels to make shot 47 match shot 46. Use the AI for the boring 70% — global match, reference starting point, log normalization — and use your panel, your trained eye, and your finishing room for the part that actually justifies your rate. Treat it like an assistant that does the prep so you can do the work.

---

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