Color Grading Portraits with AI: A Practical Workflow

The look I've been chasing for about three years now is straightforward to describe and surprisingly hard to land consistently across a wedding gallery: warm skin, slightly cool shadows, an honest highlight that does not blow out, and a clean neutral midtone that lets the rest of the image breathe.

This is not a "moody dark editorial" preset, and it is not a "bright and airy" preset either. It is, more or less, a print-faithful color grade, with the small amount of personality I've decided suits a wedding. Getting there used to take me twenty minutes a frame for a portrait, and an hour per gallery for the catalog passes. AI has shortened both numbers, but the more interesting thing is what it has changed about how I work, not just how long.

What I'm color grading for

A wedding gallery is not a single image. It is, on average, six hundred to nine hundred photographs, taken over fourteen hours, under at least eight distinct lighting conditions. The color grading job, on a wedding, is really two jobs:

• Per-frame correction. White balance, exposure, individual color casts. Mostly invisible when done well.
• Gallery cohesion. Making sure that the ceremony, the reception, the getting-ready light, and the outdoor portraits all sit inside a single visual language when a viewer scrolls through the gallery.

If you only solve the first problem, you get a clean but unsigned gallery — competent and forgettable. If you only solve the second, you get a moody gallery where every other frame looks subtly wrong. The work, and the place where AI now helps me most, is doing both at once.

Step one: get the white balance honest

I shoot with a custom white balance set at the start of every scene, and I bracket my reference frame with a gray card whenever I have a quiet second. This is not glamorous, and it is the single most consequential thing I do for the rest of the pipeline. If your starting white balance is off by 400 Kelvin across a hundred ceremony frames, no amount of AI cleverness later will save the skin.

In Lightroom, I use the auto-WB AI suggestion as a sanity check, not as a default. About seventy percent of the time it agrees with my reference frame; the other thirty percent it's leaning a little cool, and I trust the gray card.

The AI is not deciding the color. It's offering a second opinion on the math.

Step two: build the base grade as a profile, not a preset

I stopped using presets a few years ago for color grading and switched almost entirely to camera profiles plus targeted HSL. The reason is consistency: a preset moves your entire image somewhere, and a profile changes how the raw data is interpreted before any sliders touch it. For a multi-frame gallery, profile-based grading is far more stable.

The base profile I use is a custom-built DCP, modeled off my film references and applied per camera body. Building one used to require Adobe's DNG Profile Editor and a patient afternoon with a color chart. Building one in 2026 still requires the same patience, but the iteration loop is much faster because AI denoise and color tools let you see the result on actual real-world frames almost in real time.

The targeted HSL pass

On top of the profile, I do a light HSL pass: pulling oranges (skin) toward a slightly warmer hue, pushing the reds in florals away from the skin tone, knocking the saturation off magentas and yellows by about ten points each. This is the same pass I have done for years. Nothing AI here. Just taste.

Step three: use AI masks to do work the global sliders cannot

This is where the workflow has changed the most in the last two years. The AI masks in Lightroom — Subject, People, Sky — allow me to do targeted color work that used to require Photoshop round trips.

On a typical portrait, I'll do three masked moves:

1. Skin mask: a one-stop lift in luminance, +5 warmth, -3 saturation on red. The skin reads cleaner and warmer without making the rest of the frame ochre.
2. Sky or background mask: -10 saturation, -5 luminance. Pulls the background back and lets the subject sit forward.
3. Subject mask, soft contrast curve: a gentle S-curve applied only to the subject, leaving the background graded separately.

None of this is exotic. What's new is that the masking is now reliable enough to apply across a whole gallery via the sync function. The bride's skin in the morning getting-ready frames and the bride's skin under tungsten reception light are now color-corrected with the same mask logic, and the result is a gallery that feels visually unified without me hand-painting every frame.

Step four: catalog-wide cohesion

After per-frame work, I do a final pass at the gallery level. I scroll through the whole take at small thumbnail size, looking for visual jumps — a single magenta frame, a too-warm interior, a noticeably cooler outdoor block. The eye picks these up at speed in ways it can't at full size.

For any frame that fights the rest of the gallery, I'll either pull it back into the family or pull it out of the delivery entirely. Cohesion is more important than salvaging a frame that wants to live in a different gallery. Color grading is, partly, also editing.

Where I do not use AI for color

For sake of honesty:

• I don't use "auto-grade my whole gallery" features. Even when the result is technically competent, it tends to be visually average. Average is the enemy of memorable.
• I don't use AI-generated LUTs from prompt-style tools for client work. Useful for personal experiments. Not stable enough across mixed lighting for a delivered wedding.
• I don't use AI skin-tone "correction" that aggressively renormalizes skin to a target value. Real skin has variation. The point of the portrait is to show that variation, not to flatten it. The BLS occupational outlook talks about photography as a craft of "capturing subjects with personality and detail," which I think is exactly right — and skin tone is where that personality first lives.

The grade is the second draft of the photograph

I tell new photographers something that took me a while to internalize: the grade is not where the photograph is made. The photograph is made in the room, in the moment, with the people. The grade is where you finish the sentence. AI can make the finishing faster. It cannot make the sentence.

That distinction has shaped how I think about all of these tools. Use them aggressively for the parts of the work that are mechanical. Hold them at arm's length from the parts that require taste. The result, in my workflow, is more time spent looking at images and less time spent fighting them — which is what I wanted from these tools in the first place.

If you want the bigger-picture take on where AI is taking portrait work, the first piece in this series is the longer read. Or if you're after the rest of my production stack — culling, masking, denoise — the tools roundup picks up where this piece stops.