Inspiration
The idea came from one simple, uncomfortable question: What happens when A.I. characters understand they’re A.I.? Turning that into a child interview gives it emotional weight you don’t expect. It’s uncanny and human and deeply weird in the best way. I wanted to create this as an experiment
How we built it
We started with a single base prompt that defined tone, age, and emotional intention. Then we let the model improvise. No scripted answers. No pre-written dialogue. Just guidance + reasoning. After that, it was about refining looks, stabilizing motion, replacing weak frames, and shaping the final cut into something that feels like documentary realism
Challenges we ran into
Actually not alot to be honest! Sometimes the model was a bit too restrictive so a few generations failed.
Accomplishments that we're proud of
The emotional authenticity. The coherence of answers despite no script. The sense that the child knows more than it should. And of course: the fact the whole thing feels disturbingly real.
What we learned
Guiding reasoning models is the new directing. It’s not only prompting anymore. Letting the model think and then shaping how it thinks. That’s a new craft.
Built With
- kling
- nanobanana
- sora2
Log in or sign up for Devpost to join the conversation.