SIGGRAPH 2026 — wrap-up and thoughts
“The winning tools treat the artist as the author and the model as the assistant — never the other way around.”
The show floor finally stopped arguing about whether the tools work and started arguing about how to use them well. That is real progress.
I have been walking these halls since the days when a real-time preview was a research paper, not a product. This year felt different. The panic has burned off. What is left is craft.
Two years ago every booth promised to replace you. This year the good ones promised to get out of your way. Denoisers that respect your art direction. Solvers that expose their guts so a TD can actually debug them. Tools built by people who have shipped a show.
The best demo I saw all week was not a generator. It was a lookdev tool that let an artist paint intent and had the model fill the tedious middle, with every decision inspectable and reversible. That is the shape of the future I want.
I flew home more optimistic than I have been in a decade. The work is changing. The reason we do it has not.
Why the word "AI" isn't descriptive enough any longer
Calling everything "AI" is like calling every tool a "computer." True in 1985, useless on set today. We need better words.
A producer asked me last week whether a shot was "done with AI." I did not know how to answer. Which part? The rotoscope assist? The generative fill on a wire removal? The layout blocking a model suggested that we then hand-animated over?
The word has become a bucket so large it stopped meaning anything. And when a word stops meaning anything, people project their fears into it.
Comparing modern path tracers from a hardware perspective
Everyone benchmarks samples-per-second. Almost nobody benchmarks samples-per-watt, or how gracefully a renderer degrades when the scene will not fit in VRAM.
I spent a month with three production path tracers and a rack of very different hardware. The headline numbers you see in marketing decks are real, and also close to useless for planning a facility.
For a studio, the question is not "how fast is one frame on the fastest card." It is "how many artist-hours and watt-hours does a sequence cost from first light to final." Those are different questions, and they favor different renderers.