How to Fix Multi-Cam Shot Matching (Indie) | Leumos AI

Multi-camera shot matching breaks on indie sets when a BMPCC 6K, Canon C70, and Sony FX6 share a scene with different sensors and log curves. The fix is a three-pass workflow: input color space normalization first, AI-driven equalization across the timeline second, then surgical primaries on the holdout shots. Done right, the first equalization pass becomes one automated match plus a review round instead of a clip-by-clip crawl.

I've been grading indie shorts and low-budget features for years, and the pattern is identical on almost every project. The DP shoots a BMPCC 6K main and a Canon C70 B-cam because the rental package only had one of each. Then someone shows up on day three with a Sony FX6 for the gimbal shots. Now I'm staring at 200 clips across three sensors, two log curves, and a director who wants it to look like Aftersun. The first day of the grade used to disappear into manual equalization before I touched a single creative decision. This piece is the workflow I've built to keep that from happening.

Why indie multi-cam grades break in the first place

Three sensors mean three color sciences. BRAW out of a Pocket 6K leans cool and contrasty with a punchy magenta cast in skin. C-Log3 from the C70 sits flatter and pulls warm — Canon's color science has always favored a slight orange-yellow bias in midtones. S-Log3 on the FX6, mapped to S-Gamut3.Cine, renders highlight rolloff more graciously than either of the others and reads skin a touch cleaner — assuming you expose S-Log3 correctly in the first place. RED Komodo's R3D adds a fourth dialect — IPP2 with REDWideGamutRGB is its own conversation entirely.

Stack those clips on a timeline and even a perfect on-set white balance won't save you. The DIT's gray-card pass gets you within striking distance, not all the way home. You still have to reconcile the rendering of midtones, the saturation falloff in highlights, and the specific way each sensor handles the practicals in a coffee-shop interior. On a $20K short, you don't have a colorist and an editor — you're both. That doubles every minute spent on the equalization pass.

This is the bottleneck that AI shot matching for color grading can actually solve, and it's the one most worth solving.

The three-pass workflow: normalize, equalize, sculpt

The pipeline I've landed on has three discrete passes, and they have to happen in this order.

Pass one: normalize the input color space. Every clip gets converted from its native log gamma to a common working space — Rec.709 for offline, or DaVinci Wide Gamut if you're roundtripping back to Resolve for the finish. This is the cleanest place to apply Blackmagic's BRAW conversion, Canon's official C-Log3 to Rec.709 LUT, Sony's S-Log3 to S-Gamut3.Cine LUT, and RED's IPP2 output transform. Don't eyeball this. Use the manufacturer's transforms.

Pass two: equalize across the timeline. This is where you reconcile the exposure drift between cameras, the temperature bias each sensor introduces, and the saturation differential. Doing this clip-by-clip with Resolve's built-in shot match feature works, but it's slow, and shot match struggles when one of the source clips is itself badly exposed. Automatic shot matching that analyzes the whole timeline at once does better here because it has more reference data to average against.

Pass three: sculpt the look. Only now do you make creative decisions. A teal-orange Sayombhu Mukdeeprom-style warmth like Call Me By Your Name, a desaturated washed grade in the Chayse Irvin mode on Aftersun, a heavier silver-retention pull — whatever the film is asking for. If you've done the first two passes correctly, this is the fun pass and it goes fast.

Most colorists I know used to do all three of these passes simultaneously and wonder why every grade took twice as long as it should. Separating them is the unlock.

If you're an indie filmmaker, we're building this for you. Leumos AI is in closed beta — join the early-access list and you'll be in the first 500 (50% off for life).

Where AI shot matching actually shines (and where it fails)

I'll be honest about this because the audience for this page already knows the limits. AI shot matching is excellent at equalization passes — flattening exposure variance, neutralizing temperature drift, harmonizing saturation across mixed-sensor footage. On a coverage scene with consistent lighting and three cameras pointed at the same talent, a good AI pass will get you 85-90% of the way to a matched timeline.

It fails predictably in three places.

Mixed lighting. A reception scene with tungsten practicals, daylight through a window, and an LED panel hidden behind the cake — no AI tool currently on the market handles this cleanly. The math can't decide which light source to trust. You'll need to mask and grade manually.

Skin in shadow transitions. When the talent walks from a key-lit area into deep shadow within the same clip, AI matching tends to over-correct the shadow exposure and crush the highlights. Manual sculpt.

Creative grade intent. AI doesn't know the director wants a slightly green undertone for the third act because the protagonist is sick. That's still on you.

The honest framing is this: AI shot matching takes over the equalization pass and does nothing at all for the creative grade. If you treat it as the former, it earns its place in the pipeline. If you expect the latter, you'll be disappointed.

Tools worth knowing in this space — Colourlab AI does aggressive look-matching with strong DIT-pipeline integration. fylm.ai has a clean browser interface and solid reference image matching. Dehancer leans into film emulation harder than anyone. Each has real strengths. I'm building Leumos because none of them currently solve the specific bottleneck I keep hitting on indie features: a fast browser-based first pass that hands off cleanly to Resolve without forcing me to switch grading philosophies.

Building the pipeline: browser-based first pass, Resolve finish

Here's how the workflow will sit when Leumos opens for early access.

Step one: drop your full timeline export into the browser, whatever it was shot on. AI Scene Cut Detection chops it into individual shots automatically. You don't build a node per clip the way you would in a fresh Resolve project.

Step two: apply the Input Color Space LUT for each camera. One click per source — S-Log3 to Rec.709, C-Log3 to Rec.709, BMDFilm, V-Log if you've got a Panasonic in the mix. Now everything is sitting in the same working space.

Step three: hit Match All. The AI equalizes exposure, contrast, brightness, saturation, and hue across the entire timeline in a single pass. This is the step that used to be a clip-by-clip crawl through the whole timeline.

Step four: drop a reference still — a frame from Hoyte van Hoytema's grade on Oppenheimer if you're cutting B&W sequences, or a Rachel Morrison still from Mudbound for a desaturated naturalistic look — into Reference Image Grading. The intensity slider lets you dial the influence so you're not getting a slavish copy.

Step five: clean up the holdouts. Any shot the AI mis-matched gets fixed with Manual Primaries — exposure, contrast, white balance, saturation. Use Manual Cut Tool for any shot boundary the AI missed. The Preset LUT Library is there if you want to layer a baseline emulation underneath the match.

Step six: export the graded render and drop back into Resolve for the finish — secondary qualifiers, power windows, film grain, deliverables — using the render as your approved reference. The creative grade happens where you already have the most powerful tools.

The point isn't to replace Resolve. The point is to stop hand-walking the front of a grade through a problem that is purely mechanical.

What to do when the cameras truly don't match

Sometimes the footage is genuinely irreconcilable. A drone shot with crushed shadows because the operator forgot to expose for the highlights. A handheld clip shot at the wrong color temperature for forty minutes. An interview shot on a borrowed camera with a different log curve than what was on the call sheet.

When that happens, don't fight the equalization pass. Pull the offending shots out of the AI match, grade them separately, then re-integrate. The trap is trying to force one global match across footage that has fundamentally different exposure information — you'll end up degrading the good shots to meet the bad ones.

The 80/20 rule for indie multi-cam: AI matches 80% of your shots and hands you back the other 20%. Those holdouts are where nearly all of your hands-on time goes, and that's normal. Plan for it.

For festival deliverables — Sundance, SXSW, TIFF — programmers won't notice individual clip-by-clip matching errors unless they're catastrophic. They will notice if the film feels visually inconsistent across acts. That's the bar. Focus your manual sculpt time on act-to-act cohesion, not pixel-perfect clip matching.

I'm building Leumos to be the browser-based front of this pipeline. Once early access opens, the first 500 signups lock in 50% off for life. Plan details will be announced closer to launch. Join the early-access list and I'll send you the launch link the day it goes live.

Frequently asked questions

Can AI shot matching reliably handle footage originated on BRAW and ProRes RAW in the same timeline?

Yes, once both are exported out of their raw containers and normalized to a common working space. The trick is doing the input color space transform first — Blackmagic material carries BMDFilm gamma, RED material carries Log3G10, and they need different transforms. Apply the correct one per source, then run the AI match. Skip the normalization step and the AI is trying to equalize footage that's still in different color spaces, which produces inconsistent results. Once everything is normalized to one working space, the first automated pass does the bulk of the work and you hand-finish the shots that matter.

Should I bake the input color space transform into the file before grading?

No, keep it as a LUT or transform node, not a bake. Baking destroys the highlight latitude you're paying for when you shoot RAW in the first place. If a director changes their mind about the look — switching from a Mudbound-style desaturated grade to something warmer like Call Me By Your Name — you want to roll the transform back and re-grade, not re-conform from camera originals. The exception is offline editorial: bake a Rec.709 proxy for the edit, keep the RAW originals untouched for the grade, conform back for the finish. Standard indie workflow.

How do I handle skin tones across BMPCC, Canon, and Sony in the same scene?

Skin is the place AI matching is weakest because each sensor renders melanin differently and the math can't always reconcile that. My approach: let the AI equalize the scene globally, then pull a secondary qualifier on skin tones across all three cameras and apply a small uniform shift — usually a slight desaturation and a 5-10 point hue nudge toward orange to anchor everything. Sayombhu Mukdeeprom does something similar on Call Me By Your Name where skin reads warm but never plasticky. The AI gets you 80% there; the qualifier gets you the rest.

Is browser-based color grading fast enough for a 90-minute feature?

For the equalization first pass, yes — that's specifically what the architecture is built for. You're not doing finishing-grade work in the browser. You're doing the time-killer that comes before the creative grade. A 90-minute feature with ~800 clips runs the AI scene cut and Match All passes without you sitting through them. The finishing grade — secondary qualifiers, power windows, tracking, film grain, deliverables — still happens in Resolve where it belongs. The browser is the front of the pipeline, not the whole pipeline.

Will festival programmers or colorist supervisors accept an AI-assisted grade?

They already do — most of them just don't know it. Resolve's built-in shot match is AI-assisted. Colourlab AI has been in features that played Sundance and TIFF. The supervising colorist on a finishing house may want to know what was used in the offline pass, but the question is always whether the final grade holds up creatively, not which tool got you there. Festival programmers care about visual cohesion act-to-act and emotional tone. They don't care whether a Match All button handled your equalization pass or you walked it clip by clip.

What's the workflow for getting from a browser-based first pass back into Resolve?

Export the graded render from the browser, then conform back to your Resolve project with that render as your approved reference. Leumos hands back rendered files rather than grade data, so the match is a target you're matching to, not a node stack you inherit. From there you build secondaries, power windows, qualifiers, film emulation, and deliverables the way you normally would. Nothing about your Resolve finishing setup changes. The browser pass replaces the stretch you used to spend before you ever touched a creative decision. The finishing work happens where you have the most powerful tools.

How does AI shot matching compare to Resolve's built-in shot match feature?

Resolve's shot match works on a pair-wise basis — you pick a reference clip and match a target clip to it, one at a time. It's accurate but slow, and it breaks down when the reference itself is poorly exposed. AI-based timeline matching analyzes every shot in the project simultaneously and reconciles toward a global midpoint, which produces more cohesive results across mixed-camera footage. They're not in competition — I use both. AI for the first equalization pass on the full timeline, Resolve's shot match for surgical fixes on the holdouts that didn't land cleanly.

---

Leumos AI is in closed beta. The first 500 waitlist signups lock in 50% off for life. Join the early-access list →