Summary: 5 of 100 random clips were good enough; 8 of 20 after better prompts; a blind check kept 12 of 20.

Our movies: quality, not quantity

Concept and project videos made by our own AI tools. Updated 15 Oct 2026.

What they are

Short clips from our own video model, 3D scenes made in Blender, recordings of our live sites with captions, and pictures from our image model. We make a clip only from a written brief; the endless clip-making was stopped on 13 Oct.

How they are judged

What changed

At first only 5 of 100 random clips were good enough. With better prompts (one subject, visible motion, no screens) it rose to 8 of 20. The main faults left are garbled text, warping and flicker.

What they are used for

Project sites and promos, product concept films, the daily summary video, and one best piece an hour in the team room. Only clips that passed the judge.

Next: the lessons go into the standing prompt, and the blind check runs every week.