LECINEMA JOURNAL

Notes on making and watching AI film

Model tests, festival news and the stories behind the short films we show.

The same close-up of the bike mechanic from the text-only test, side by side: Gemini Omni 1.2 on the left and Wan 3.0 on the right, each pushing up his glasses.

Article

Gemini Omni 1.2 vs Wan 3.0: a blind AI video test Nov 2026

One 15-second, five-shot scene, two AI video models, two takes each, with and without reference images. A blind AI judge preferred Gemini Omni 1.2 for following the script, and it costs about 27% less per clip. Watching the clips myself, I found Wan 3.0 kept the location more consistent when the camera changed angle. Here are the scores, the frames and which to pick when.

By Andy Ratsirarson8 min read

Read the article

New tests and festival notes land here as we publish them.

Try it in LeCinema Studio

Write a scene, generate the shots with AI video models and cut them into a short film.