Film editing happens at two scales. Macro edits — which scenes go where, how the story is structured — get most of the attention in editing tools. Micro-edits — nudging a cut a few frames earlier to land on a character’s reaction, stretching a meaningful silence by a beat, adjusting exactly where a shot ends to redirect the viewer’s eye — are slower and more tedious precisely because standard tools make editors translate a high-level narrative goal into frame-by-frame manual adjustment by hand, every time.
Micro-Edits, from researchers at Cornell and Adobe Research, starts from a formative study with professional editors to understand how these small adjustments actually get made, then derives a mathematical formalization of that structure — treating micro-edits as continuous optimization within audio-visual synchrony-constrained windows, where audio and video can be safely decoupled and adjusted independently without breaking sync. On top of that formalization, the team built tools that use AI and optimization to balance character focus against visual cues like cut-on-action, cut-before-blink and reaction-shot timing, validated through multiple user studies against real footage.
The paper earned a UIST 2026 Best Paper Honorable Mention — not a future-dated preprint, but reviewed, presented, and recognized work, which is a rarer thing to find in this beat than a flashy unreviewed arXiv drop. For anyone building creative tooling, the project is a clean case study in a pattern worth watching generally: the most tedious parts of a craft are often the most structured parts, once someone bothers to formalize what a practitioner is actually optimizing for by feel.