Motion studies
AI production range
These are working tests. Each one is after a different problem: a character that holds across shots, voice and lip sync, camera movement, lighting that doesn't drift. Some of the clips are rough. That's the point. The fastest way I've found to get better at motion generation is to break things on purpose and take notes.
01 / 06
Episode 1
Maika Moving In
This was me learning how to make UGC video ads. I wanted the whole run, not just a generated clip: the process, the pipeline, which assets you need in hand, and a script that can actually carry the spot.
The result is a realistic influencer character living her life in a surf town, mixed with a short product sell. The lifestyle has to hold first. The sell has to sit inside that life, not land like a separate ad at the end.
I don't run these as a single prompt-and-generate call. Production goes through node-based workflows: reference images, audio, and staged prompts chained across steps. The clips above were built that way in Seedance and Seedream. Some are single shots. Some are longer sequences with synced dialogue.
I add more of these as the tools change. Treat this page as a running log, not a finished set.