Automatic shot matching can shorten the repetitive part of an indie grade, but the useful unit of work is still the shot. Normalize the source correctly, match a selected shot to a reference, check the result in motion, and make a deliberate decision before moving to the next cut.
That is also the current Leumos workflow. Reference matching runs on one selected shot at a time. Leumos does not automatically reference-match a full timeline in this release. It detects cuts, keeps the shots in one timeline, gives you manual and AI-assisted grading controls, and exports the graded timeline when the review is complete.
The Math of Indie Multi-Cam Is Brutal
Mixed-camera footage creates two different problems. The first is technical: each camera may encode tone and color in a different log curve and gamut. The second is photographic: lenses, exposure, white balance, lighting and production design can change from shot to shot even when the cameras are identical.
Solve the technical problem first. Choose the input transform that matches the actual camera setting, such as the precise S-Log3 gamut or C-Log3 combination used on set. A transform is not a creative look. It gives the shots a common starting space so that a comparison is meaningful. Our guide to CST, ACES and LUT differences explains why two plausible transforms can still produce different pictures.
Then group shots by scene and lighting condition. A daylight exterior and a tungsten interior should not be forced toward the same neutral average. Within each group, choose a representative shot with a clear subject, useful highlight and shadow detail, and no temporary lighting effect. That shot becomes the visual reference for the group.
Scene cut detection helps turn a finished edit into a shot timeline. Treat its boundaries as an editable first pass. Dissolves, flashes and fast camera moves deserve a quick check before any grade is copied or matched.
What Manual Shot Matching Costs You Creatively
The cost of manual matching is attention. Repeating the same exposure and white-balance checks across a long timeline makes it easier to miss the cut that actually needs judgment: a face moving into mixed light, a practical that should stay warm, or a highlight that is clipped in only one camera angle.
A review order keeps the routine checks from swallowing the creative pass:
- Confirm the input transform and levels.
- Match exposure and color balance within one scene.
- Check skin separately from the overall image.
- Play through the cut and look for a visible jump.
- Decide whether the shot should match or remain different for story reasons.
This order also makes automation easier to supervise. A matched frame can look convincing while the moving shot reveals a bad mask, a changing light source or a clipped channel. The answer comes from playback and scopes together, not from a confidence percentage.
Leumos keeps those checks in the Studio. You can inspect exposure with the waveform and vectorscope, refine the overall image with the grading controls, and work on skin with a dedicated secondary. A more advanced finishing application may still be the right place for arbitrary HSL qualifiers, tracked windows, a raw conform or a node-based delivery pipeline.
How AI Shot Matching Actually Works Under the Hood
For an editor, the most useful explanation is the sequence of decisions rather than a promise about the model.
First, the input transform maps the camera source into the working display space. Next, scene detection identifies likely cut points. Reference matching then analyzes the selected shot and still image, produces a proposed grade, and returns it for review. The proposal is applied to that selected shot. It is not propagated automatically across the timeline.
Review the match in motion before accepting it. The waveform shows whether shadows, midtones and highlights moved into a sensible range. The vectorscope shows color balance and saturation without relying on a laptop display alone. Leumos also has a dedicated Skin tone workflow: automatically select skin, inspect the selection mask, refine it with Select, Add, Remove, Range and Softness, then adjust hue, warmth, color boost, brightness and contrast. This is a skin-specific secondary, not a general HSL qualifier or a claim of automatic subject tracking.
Use this checklist on every representative shot. It is a review aid, not a benchmark or an accuracy score.
| Check | What to inspect | Decision |
|---|---|---|
| Input | Camera log curve, gamut and levels | Choose the exact transform before judging the look |
| Exposure | Meaningful blacks, midtone placement and highlight detail on the waveform | Correct a technical mismatch before adding contrast |
| Balance | Neutral objects and RGB/CbCr trace relationships | Keep intentional warmth; remove accidental casts |
| Saturation | Trace spread and any channel pushed to an edge | Reduce a global push before reaching for a local fix |
| Skin | Face appearance, vectorscope direction and the visible skin mask | Refine the selection before changing skin controls |
| Continuity | Playback through the incoming and outgoing cuts | Keep the match only if the cut feels coherent in motion |
| Exception | Story lighting, flash frames, silhouettes and mixed sources | Grade separately when matching would erase intent |
Similar shots can share a grade as a starting point, but each recipient still needs the same review. Copying settings does not prove that exposure, white balance or skin will land correctly on a different camera or lighting setup.
Where AI Shot Matching Breaks (Be Honest)
Mixed lighting is the common failure case. A face can be lit by daylight on one side and a tungsten practical on the other. A global white-balance change can improve one side while damaging the other. Use the Skin tone selection and mask to see what the local correction will touch, and keep the creative lighting when neutralizing it would flatten the scene.
Clipped or missing source information is another hard limit. Matching cannot restore a channel clipped during capture or detail removed by an earlier transform. Check the untouched source and input transform before trying to repair the result with more contrast or saturation.
Stylized shots also need judgment. Day-for-night work, colored stage light, silhouettes and motivated color casts are not errors just because they differ from the median shot. A consistent timeline can contain deliberate differences.
Finally, a still image cannot show temporal problems. Play the whole shot and the adjacent cuts. Watch faces, practicals and bright edges as they move. Leumos provides a visible skin mask for inspection, but this article makes no claim that a mask will follow every subject automatically through a shot.
There is no responsible universal percentage for how much of a grade AI completes. The result depends on capture, transforms, reference choice, lighting and the standard of the delivery. The checklist above makes those variables inspectable.
The Resolve-Friendly Workflow I'm Building For
Leumos is live and works well as a focused stage inside an existing post-production workflow:
- Finish the edit in your NLE and keep a separate camera-original conform if the delivery requires one.
- Export a file the browser workflow can read, without baking an unintended display LUT into log footage.
- Import it into Leumos and inspect the detected shot boundaries.
- Choose the correct input transform for the source.
- Match one selected shot to the reference and review it in motion.
- Check waveform, vectorscope and the Skin tone mask; refine the shot with manual controls where needed.
- Repeat the review for the remaining shots, using a copied grade only as a starting point for genuinely similar material.
- Export the graded timeline, or return to Resolve for any raw conform, arbitrary qualifier, tracked window, node-based finish or delivery requirement outside Leumos.
Leumos returns rendered video rather than Resolve nodes, XML, CDL or a LUT. Export resolution and bit depth depend on the plan, and Studio includes a ProRes 422 HQ option. The current options are listed in pricing.
The privacy boundary changes at export. Opening a clip keeps playback, scrubbing and most color work on the computer, while scene detection and reference matching send the reduced copies or frames described in the policy. Export uploads the complete original file for cloud rendering and provides a time-limited download link. Read the exact transfer cases in What leaves your computer, and when.
Frequently asked questions
How long does automatic shot matching take on an 80-minute indie feature?
There is no supported fixed estimate. Leumos reference-matches one selected shot at a time, so total working time depends on the number of shots you choose to match, upload speed, source format, render queue and how much manual review each shot needs. Automatic full-timeline reference matching is not available in this release.
Does AI shot matching work with footage from BMPCC, Sony, and Canon bodies in the same timeline?
The shots can live in one timeline, and Leumos includes named input transforms for common Blackmagic, Sony and Canon log/gamut combinations. Choose the transform that matches each source, then review shots individually. Converting different cameras to a common space removes one source of mismatch; it does not make their exposure, lenses, lighting or sensor response identical.
Can I round-trip my Leumos grade back into DaVinci Resolve for final finish?
Leumos exports rendered video, not Resolve grade data. Use the render directly when it meets the delivery requirement, or use it as a visual reference during a separate conform and finishing pass in Resolve. Leumos does not export its grade as nodes, XML, CDL or a LUT.
Will AI shot matching get skin tones right across different lighting setups?
Do not rely on a percentage. Automatically select skin, inspect the mask, refine the selection, and compare the result on the vectorscope and in motion. Mixed practicals and changing light still need judgment. Leumos supplies a dedicated skin secondary; it does not replace every qualifier or tracked correction in a finishing application.
What's the difference between Match All and Reference Image Grading?
Reference Image Grading proposes a look for one selected shot using a still image. An automatic Match All pass across the full timeline is not available in this release. For similar shots, copy a grade as a starting point and review each destination shot.
Is this a DaVinci Resolve replacement?
Leumos is a browser color-grading studio with scene detection, selected-shot reference matching, manual controls, waveform, vectorscope, a dedicated Skin tone secondary and rendered export. Resolve remains the broader tool for editing, audio, raw conform, arbitrary qualifiers, tracked windows, node-based finishing and delivery formats outside the current Leumos plan.
How does the pricing work for an indie feature with a 4-week post window?
Plans differ by file limits, monthly reference matches, export resolution and bit depth, and cloud-footage allowance. Check the live pricing section for the current limits before choosing a plan.
Leumos is open. Match a selected shot to a reference, inspect the result with scopes, refine it, and export according to your plan. Start in your browser