CG/VFX Supervisor. Brand films, ads and product motion built with generative AI, then finished with CGI, compositing, edit, sound and grade.

A cut of recent AI production work — brand films, commercial spots, product motion, motion graphics and short-form social. Every shot in it was generated, then taken through the same finishing process a conventional job gets.

I came to AI production out of 20+ years of broadcast television and feature film VFX and CGI. That is the standard the work is still held to.

No single model carries a whole job. Each one holds different things well — camera behavior, physical motion, human performance, clip length, how closely it follows what it is told — and each one fails in its own way. Shots are routed to the model that holds that particular brief, which means running several generation systems in parallel and knowing their behavior well enough to call it before the render rather than after.

Prompt-based systems take the shots that can be described. Node-based work takes what description alone can't reach: reference conditioning to lock identity, first- and last-frame control, depth and edge maps driving motion, prompt scheduling across a shot, interpolation and upscaling on the way out.

Continuity is built, not hoped for. Identity language is held fixed while camera and scene language changes, frames are chained out of approved shots to seed the next, and lighting and lens carry across a sequence.

Product accuracy, logos, type, tracking and clean-up are built and composited the traditional way. The cut, the sound and the grade go through the same process as any broadcast job.

Product motion is where generative AI is least forgiving. A brand will accept an invented environment; it will not accept an invented product. So the world around the product — environment, light, atmosphere, motion — comes out of generation, and the product itself is built and composited the traditional way, accurate down to the logo and the type. It reads as a product film rather than an AI experiment. Motion graphics, type and logo work sit on top of that, animated and tracked the same way they would be on a broadcast spot.

ComfyUI, Runway, Kling, Seedance, Luma, Midjourney, Higgsfield, Flora, Nano Banana, Topaz and Magnific for generation and enhancement. Unreal Engine, After Effects, Premiere Pro and DaVinci Resolve for finishing. ElevenLabs and Udio for voice and music.