Reference-Based Color Grading for Wedding Videography: Drop a Still, Match the Look | Leumos AI

I've been a DaVinci Resolve Certified colourist for four years, and the call I dread most is the wedding filmmaker on a Friday night who just shot 14 hours across a Sony A7 IV, an FX30, and a borrowed Canon R6 II, and wants the final film to look like Eric Floberg directed it. Three cameras. Two log profiles. A reception lit with tungsten bulbs, an iPhone flashlight from the best man, and the last of the golden hour bleeding through a barn door. Every shot is different. The couple expects a highlight reel that feels like a film. The turnaround is six weeks because there are eleven other weddings in the queue.

This is the math of modern wedding videography. And it's why reference-based color grading — the idea that you drop a still you love and the software pulls your footage toward it — has gone from a novelty to the only sane way to keep up.

Why LUTs alone stopped working for weddings

LUTs are static. A LUT built for an A7 IV at f/1.8 on a sunlit ceremony does nothing kind to the same camera shooting ISO 6400 under a string of warm LEDs four hours later. You end up either grading every clip by hand or accepting that half your timeline reads wrong. Most wedding shooters I know own 30+ LUT packs and use maybe four of them, because the rest don't survive a real scene.

Reference-based grading flips the question. Instead of asking "what transformation matches this camera profile," it asks "what does this still look like, and how do I pull this clip toward it?" The reference is the destination. The footage is the starting point. The software handles the math.

For wedding work, that's a meaningful shift. You can look at Eric Floberg's editorial wedding films — those cool, slightly desaturated skin tones, that lifted shadow that never turns muddy — and use a still from one as your guide rather than reverse-engineering the look with primaries and curves. You can pull a frame from a Stories by Joseph Radhik film, the kind of warm, gold-laced palette that defines premium Indian wedding cinematography, and aim at it.

How reference-based grading actually works

The underlying process is histogram and color distribution matching, augmented with machine learning on what "skin" looks like so the algorithm doesn't push a bride's face toward chartreuse trying to nail a backdrop. Good implementations read three things:

• The luminance distribution of the reference (where shadows, midtones, and highlights sit)
• The chroma channels (the actual color tilt — warm, cool, magenta-shifted, green-pulled)
• Skin tone protection (so the match doesn't murder faces)

It's not magic. It's a directed contrast and color transform with intelligent guardrails. The reason it works for weddings specifically is that wedding stills already exist in abundance. Couples send them to you. You scroll Instagram and save them. Your reference library is the photo deliveries the couple already loves.

A real workflow: Sony A7 IV + FX30 + R6 II, six-week turnaround

Here's the workflow that actually saves time. I'll walk through it the way I'd do it on a real wedding.

You upload the selects — ceremony, first dance, reception, golden hour portraits. The software auto-detects scene cuts and lays them out as a timeline with thumbnails. You don't node-per-clip like in Resolve. You're looking at the day, organized.

Step one is the input color space transform. S-Log3 from the A7 IV, S-Log3 from the FX30, C-Log3 from the R6 II — all of it gets pulled to Rec.709 in one click. Now you're looking at footage in a viewable space. You'd be surprised how many wedding shooters skip this and try to grade in log; it's like painting with the lights off.

Step two is the reference. You drop in the still — say it's a Floberg-style editorial frame the couple already loved from their engagement photos — and apply the grade. Pull the intensity slider down to 70% so you're not crushing into pure imitation. The footage now feels like that world.

Step three is making the timeline cohesive. The reception clips were shot at 6400 ISO under mixed light; the ceremony was at 800 ISO under open shade. Without help, they read like two different films. The fix is timeline equalization — exposure, contrast, brightness, saturation, and hue normalized across the whole roll so the cuts breathe together. This is the unglamorous middle 60% of any wedding grade, and it's where every hour saved adds up across a season.

If you're a wedding filmmaker, 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).

Step four is the cleanup pass. You always have a few clips the AI doesn't quite catch — a wide of the reception where the chandeliers are blown out, a tracking shot through a corridor with mixed daylight and tungsten. You go in with surgical primary controls — temperature, tint, exposure — and finish them by hand. The bulk of the timeline is already there; you're polishing the 10% the algorithm couldn't read.

Where Leumos AI fits this workflow

I'm building Leumos AI because I got tired of toggling between four tools to deliver a wedding film. The MVP is browser-based and the wedding-specific workflow is the thing I obsessed over.

• Reference Image Grading lets you drop the photographer's still, a Floberg frame, or even a Lightroom export from the couple's engagement set, and pull footage toward it.
• Match All handles timeline cohesion — the multi-cam, multi-light reality of weddings — so the ceremony and the reception don't read like different productions.
• AI Scene Cut Detection chops uploads into a shot timeline automatically. No more dragging in clips one at a time.
• Input Color Space LUT handles S-Log3, C-Log3, BRAW, and V-Log → Rec.709 in one click, so the day starts in viewable space.
• Manual Primaries and the Manual Cut Tool are there for the 10% the AI doesn't catch.
• The Preset LUT Library is there when you want a starting point instead of a reference image.

The honest framing: Colourlab AI is more powerful for narrative work. Dehancer's halation and film emulation is genuinely beautiful. fylm.ai's color spaces and node-based workflow are deep. But none of them were built for the rhythm of a wedding shooter — multi-cam, multi-light, six-week turnaround, $1,500-$8,000 budgets, and a couple who wants their film to feel like the work of a photographer they already love.

Where reference-based grading still falls short

I'll be straight with you, because credibility matters more than hype.

Reference grading struggles when the reference itself contains color information the algorithm misreads as the intended look. A still with a strongly tinted background can pull your skin tones with it. Always check faces first.

It also struggles in heavily mixed light. A reception with golden hour pouring through windows, tungsten string lights overhead, and an LED on the dance floor is a three-source mess. The algorithm can land a coherent base on average, but individual frames may need manual rescue. Same for day-for-night decisions — those are creative choices, not match decisions, and no AI is going to read your mind.

Brand-Pantone work is also out of scope. If a corporate client needs a specific PMS for a logo in a brand wedding film, you're nailing that by hand. Reference grading is for the emotional palette, not the spec sheet.

What this actually saves you

A wedding film I used to grade over three days I can now get a strong base on in 90 minutes. The finishing is still mine — the burn-in moments, the highlight reel rhythm, the breath of a slow push at the altar. But the cohesion work, the technical match, the hours of curve-tweaking to make a six-hour reception feel like one continuous film — that's done by the time I sit down to do the storytelling work.

That math is the reason I'm building this. A wedding videographer running 25-40 weddings a season cannot afford three-day grades. The choice has been to deliver something flat or burn out chasing cinema. Reference-based grading, done well, is the third option — and it's the one that lets the color serve the story instead of eating your week.

Frequently asked questions

Will reference-based color grading work across multi-cam wedding footage from different camera brands?

Yes, and that's actually the case where it earns its keep. The workflow is to transform each log profile — S-Log3 from your Sony bodies, C-Log3 from the R6 II, BRAW from a Pocket 6K — to Rec.709 first, so all your footage starts in the same color space. Then the reference grade applies on top of normalized footage. The match isn't perfect across sensors (Sony skin tones and Canon skin tones have real differences), so expect to nudge a few clips by hand. But it gets you 80-90% of the way without per-camera LUT juggling, which is the bottleneck for most wedding shooters.

Can I use a wedding photographer's stills as my reference images?

This is probably the most practical use case. The photographer on the day has already nailed the emotional palette the couple loves — that's why they were hired. Pulling a few of their finished frames into your color tool as references means your film and the photo gallery read as one body of work, which couples consistently respond to. The catch: photographer edits often push contrast and saturation harder than video can sustain at 24fps. Drop the reference grade intensity to 60-75% rather than 100%, and you'll get a version that translates instead of a version that looks crunchy in motion.

How is this different from buying LUT packs from creators I follow on Instagram?

LUTs are a fixed transformation baked in once. They don't read your footage — they apply the same math whether your clip is overexposed, magenta-shifted, or in a totally different color space than the LUT expects. Reference grading reads the actual luminance and chroma of your footage and shapes the transform to land at the target. In practice this means it survives the lighting changes within a single wedding day — ceremony to cocktail hour to dance floor — without you having to switch LUTs four times. LUTs still have a place for stylistic starting points, but they aren't a complete grading workflow.

What about skin tones in heavily mixed reception lighting?

This is the honest limit of any AI color tool, mine included. When you have tungsten string lights, an LED uplighter on the bride's face, and warm daylight from a window all hitting one shot, no automated grade is going to perfectly separate the sources. The algorithm protects skin from being pushed into truly unnatural hues, but you'll usually need to finish those reception wides with manual temperature and tint adjustments. The time-saving is real on the 80% of footage that lives in one light source — the messy 20% always needs your eye. Anyone selling you otherwise is overpromising.

Do I still need DaVinci Resolve in my workflow?

If you do narrative work or commercial spots that need power windows, qualifiers, and frame-by-frame keyframing, yes — Resolve remains the deepest tool on the market and I still use it for finishing on weddings I want to push further. For the highlight reel and feature edit deliverables most wedding clients want, a browser-based reference workflow plus your NLE of choice (Premiere, FCP, Resolve's Edit page) is enough. Many wedding shooters I know are dropping Resolve from the wedding-specific part of their pipeline entirely once they have a faster grading option, and keeping it for branded or higher-budget work.

How long does it actually take to grade a full wedding this way?

On a typical 14-hour shoot with three cameras producing about 90 minutes of selects, my workflow is roughly: 10 minutes to upload and let scene detection run, 5 minutes to apply input color space transforms, 15-20 minutes to dial in the reference grade and intensity, 30 minutes for timeline equalization across ceremony/reception/portraits, and 20-30 minutes for manual cleanup on tricky clips. Call it 90 minutes for a solid base. Compare that to three days of node work in Resolve and the season math gets very different — you can take on more weddings or finally have weekends back.

What file formats and log profiles will Leumos support at launch?

The MVP handles S-Log3 from Sony bodies (A7 IV, A7S III, FX3, FX30), C-Log3 from Canon (R5, R6 II, C70), BRAW from Blackmagic Pocket cameras, and V-Log from Panasonic. Standard delivery formats — ProRes 422, H.264, H.265 — all upload cleanly. The Free tier caps uploads at 400MB per file (fine for individual selects, not for raw event masters), Creator at 1GB, and Pro at 2GB. For most wedding workflows you're grading selects rather than the full raw card, so the file size cap rarely bites. RAW workflows from Sony and Canon's newer codecs are on the post-launch roadmap.

---

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