I told the model: 5 shots, jacket only. No person. Shot 1 still came back with a full character already wearing it. Then the collar-tag shot invented text that is not on the real product. The clip is 23 seconds from a locked product photo, two frozen description blocks, chained stills, local voice, and each video clip forced to start and end on a still pair. The interesting part is not the model. It is what you refuse to let it invent. If you generate AI animation: where does the model leak a person or a logo you never asked for?
A red jacket is shown on an anime man under a spotlight, then as a rotating standalone product.
- 329 views
- 4 likes
- 0 reposts
- 2 replies
- 3 bookmarks
This video is Tú Michael | AI Animator 🇻🇳’s work, shown from their public post. See it on X.








