AI and automation
Can AI Color Grade an Entire Film?
A feature film is hundreds or thousands of shots, several sessions with the director and cinematographer, and a set of deliverables. Parts of that can be automated now. The whole thing, unsupervised, cannot. Here is why.
In short
- Grading a film is a series of jobs: conform, balance, matching, look, sessions, revisions, deliverables and QC.
- Several of those jobs can be automated or accelerated today: organization, first-pass balance, assisted matching, renders.
- Unsupervised grading fails on scale. A small error rate across a whole film still leaves many bad shots, and someone has to find them.
What grading a film actually involves
From outside, grading looks like "making the picture look good." In a DI it is a sequence of jobs, each with its own people and approvals:
- Conform and preparation. Rebuild the edit from camera originals, check every shot against the reference, set up color management for each camera, organize shots into scenes and groups.
- Primary balance. Bring each shot to a consistent, workable state.
- Look development. With the director and cinematographer, decide the look of the film and of each section.
- Scene matching. Make every scene play without jumps, keeping the changes that are intended.
- Secondaries and fixes. Skin, skies, distracting objects, beauty work, VFX shots that arrive late.
- Review sessions and revisions. Notes from director, cinematographer, producers, sometimes the studio or distributor.
- Deliverables. Often several versions: theatrical, HDR, SDR, streaming specifications, each checked.
- Quality control. Legal levels, flicker, missing shots, problems the grade revealed.
What can be automated today
| Job | Automation today | Who still decides |
|---|---|---|
| Conform, organization, groups | Good. Scripts and metadata sort shots by scene, camera and timecode. | The colorist checks the groups. |
| Primary balance | Good on most shots, with predictable failures (single-color frames, low-key scenes). | The colorist checks and fixes the failures. |
| Carrying a hero grade | Good. Copying a grade is mechanical. | The colorist grades the hero. |
| Scene matching | Partial. Similar shots match well; different set-ups need review. | The colorist, by eye, in playback. |
| Look development | Look options can be generated. | Director, cinematographer, colorist. |
| Secondaries | AI masks speed up isolation a lot. | The colorist decides what to change. |
| Renders and versions | Fully scriptable. | The colorist approves the output. |
| QC | Automated checks find technical problems. | People judge the creative ones. |
Blackmagic's 21.1 release describes its AI assistant support in the same spirit: handling repetitive project tasks so that people spend more time on creative decisions.
Why unsupervised grading breaks down
Error rates add up
Imagine a system that balances and matches 95 shots in 100 to a standard a colorist accepts. This is an illustration, not a measurement of any product. On a scene of 20 shots that is one bad shot. Across a film with a thousand shots it is fifty, spread across the film. Someone has to watch everything to find them. The review is still needed, so the honest goal is to make it faster, not to skip it.
Intent is not in the pixels
A flashback that should feel different, a character whose world gets colder, a deliberately mismatched cut, a day-for-night scene. None of this is visible to a system that only sees the pictures. The script, the director and the cinematographer carry it.
The material is messy
Real films mix cameras, archival footage, VFX shots at different stages, pickups shot months later, drone shots, phone footage. Each needs its own color management and often its own judgment.
Notes change things
Grading is a conversation. Notes in a session ("warmer," "make her stand out more," "the whole reel is too dark") change many shots at once. A useful system must accept those changes and not undo them.
What supervised automation looks like
The realistic model today is a colorist with better tools:
- The machine does the repetitive technical work: organizing, first balance, carrying grades, first-pass matching.
- It tells the colorist what it changed, and leaves approved shots alone.
- The colorist spends the saved time on the look, the difficult shots and the sessions.
- Every result is judged by the colorist, on a calibrated display, in playback.
That model fits small teams, tight schedules and long-form work where the same technical problems repeat thousands of times.
Where AutoBalance fits
AutoBalance, a DaVinci Resolve Studio panel in private beta, automates the primary balance of a scene and assists shot matching around the colorist's hero grade. After a re-edit it balances only the new cuts and leaves approved shots untouched. It does not grade a film unsupervised, and it does not create the look. Some shots still need the colorist's hand.
FAQ
Has any feature film been graded entirely by AI?
We are not aware of a verified case of a feature graded without a colorist's supervision and delivered to a professional standard. Claims of this should name the tools, the review process and who approved the result.
How long does it take to grade a feature film?
It varies widely with budget, footage, the number of deliverables and how many sessions the director needs. Automation mostly shortens the technical stages, not the creative sessions.
Will automation make grading cheaper?
It can reduce the hours spent on repetitive technical work. Whether that lowers the price or buys more creative time depends on the project.
Sources
- Blackmagic Design: DaVinci Resolve 21.1 press release, 8 September 2026
- Blackmagic Design: DaVinci Resolve Color page
Product names belong to their owners. Features change between releases; check the current documentation for your version.