Generative AI Films
The problem
Generative tools are easy to demo and hard to finish with. Nothing exposes that gap like committing to a whole film.
The approach
Make original work with no client and no brief, so the pipeline is tested against a real edit rather than a single striking frame.
What was built
Six films built end to end — concept, generation, direction, edit — across roughly 1,400 images and 1,950 video files.
The outcome
The techniques that survive here are the ones that go into client pipelines. Pendleton and Studio Wildcard run on what these films proved.
Six original short films made with generative tools between 2018 and 2026: Town of Porcelain People, Mushroom Cavalcade Trailer, The Alchemist Zoom, Ziggy Stardust, JK5, and Maze.
These are authored, not commissioned. There is no client, no brief, and no approval chain — which is the point. A generative pipeline that produces one arresting image is easy; a pipeline that holds character, lighting and tone across a whole edit is a different problem, and the only way to find its limits is to finish something.
The relationship to the client work runs one way. Techniques get proven here first — consistency across shots, direction over a model that would rather improvise, the point at which it is faster to shoot than to generate — and the ones that survive go into production pipelines like Pendleton’s accessories library and Studio Wildcard’s asset workflows.
Status of this page: written from the archive inventory. Film-by-film detail, stills and clips
are pending selection from roughly 1,400 images and 1,950 video files. Marked self-reported until
the individual films have their own verified write-ups.
The films are where the pipeline gets pushed before a client ever sees it.