FinalGrade benchmarks frontier video models on the controls filmmakers actually use. Every clip is graded blind by credentialed working professionals, one dimension at a time.
A benchmark for each department on a film set, built on the vocabulary the craft already uses and graded by its working professionals, paid at expert rates. Camera and lighting have preliminary results now; production design and costume publish in December 2026.
Win rate against the field, by craft, across seven frontier video models. Camera and Lighting show preliminary results from our first panel. Other crafts publish as each round closes.
Every model against every dimension graded so far, across camera and lighting. A single overall score hides most of this.
Preliminary, round 1, 3 graders. Score: 100 if a model's clip is always ranked first, 0 if always last. LTX and Hailuo were tested on fresh shots only. Camera results · Lighting results
FinalGrade evaluates AI the way the film industry judges its own work. We start from the real tasks each department does, from lighting a scene to taking a director's note, and test whether a model can do that work to a professional standard. That covers the shot itself, and increasingly the work around it: the environments productions build, and a model's ability to operate the software filmmakers rely on every day, from the edit to the grade. The results go in front of the crafts that define the standard.
We don't build models, and we don't take sides between labs. When a lab asks for an evaluation, the work goes to the artists on our panels: they are paid upfront, kept anonymous, and decide what they take part in. Public rankings stay public.
“A model can make a thousand shots in a minute. It still can’t tell you which one is right. That takes twenty years on set and a deep understanding of the craft.”