AI Usage

Here's a dedicated "How I Used AI" section you can drop right into the Fueler project page: How I Used AI This project was built entirely inside a generative AI pipeline — no traditional camera, no location permits, no crew. Every frame you see was directed through language. Higgsfield was used for character generation and consistency — building the hero across multiple outfit states, locking his identity (face, tattoos, earring, build) across four completely different environments and lighting conditions. Keeping one person coherent across a kitchen, a running track, a basketball court, and Times Square at dusk is one of the hardest problems in AI filmmaking right now. Higgsfield solved it. Seedance 2.0 handled motion — translating static character frames into cinematic movement. Every dribble, every sprint stride, every dance move on that Times Square plaza was directed through detailed motion prompts specifying camera angle, movement speed, shutter behavior, and physical physics. The goal was always to make it feel shot, not generated. GPT-4o was used as a creative co-director — developing the full script, writing scene-by-scene shot lists with Hollywood-level production detail, generating feature callouts and on-screen copy, and iterating on the commercial's creative direction in real time. The entire pre-production pipeline — concept, script, shot list, asset descriptions, copy — was built in Claude before a single frame was generated. The skill here isn't just knowing which buttons to press. It's knowing how to direct — how to write a prompt the way a cinematographer sets a lens, how to art-direct a scene that doesn't physically exist yet.

AI Tool Stack