I made a small paper sketch and used Gemini Omni to turn it into an animated portal video.
Here is the final result:
The sketch is intentionally simple: a tiny character and a door.
That was the main constraint. I did not want to start from a prompt-only AI clip. I wanted the video to begin with something physical: my hand, paper, pencil, and a very dumb drawing.
The useful bit was the door.
Once there is a door, every style change has a reason. The character walks in, the world changes. Pixel, neon, storybook, Japanese fantasy, plush toy, clay cutout.
Without that, it would just feel like a random AI style demo.
The rough workflow:
- Gemini Omni for the video generations
- Nano Banana Pro for visual references
- Lyria for the music
- Remotion for the final edit
Remotion mattered because AI video still needs normal editing.
Trim the slow parts. Mute bad generated audio. Put the clips in the right order. Add a small fade or flash where the model jumps. Render again.
Main lesson: do not ask the model to invent the whole thing from nothing.
Give it a tiny object, a simple action, and an edit rule.
The working video repo is here:
More soon as I keep playing with paper-to-video.