Wilderness Within · Runway Gen:48 Selected Finalist

A horror film that turns generative AI's weaknesses into its style.

Wilderness Within was written, generated, and cut in 48 hours for Runway's Gen:48 competition in September 2024, and finished as a Selected Finalist, beating 3,500 of the best AI artists in the world. The whole film runs on one creative idea: in 2024, AI video was inconsistent and warped reality, so we built something whose genre wanted exactly that.

Project
Wilderness Within
Role
Producer
Recognition
Gen:48 Selected Finalist
Built in
48 hours · Sept 2024
Runway Gen:48 · September 2024

Two and a half minutes, entirely AI, made in a weekend.

Wilderness Within follows a young woman searching for her sister, who vanishes on a hike in the Pacific Northwest while a shape-shifting presence closes in. There is no live footage in it. Every frame was generated, and a small team wrote, produced, scored, and locked the film inside the competition's 48-hour clock.

Wilderness Within, AI found-footage horror short
Wilderness Within. Generative AI short film · Runway Gen:48 Selected Finalist · 2024
The thesis

We made AI's limitations the aesthetic.

In 2024, generative video had real and obvious weaknesses. It could not hold things consistent from shot to shot, and it warped and melted reality in ways you could not fully direct. Instead of fighting that, we picked a genre that wanted it. Surreal horror runs on instability and dread, so the exact failure modes that would sink a clean commercial were the thing that made this feel unsettling and alive. The weakness became the mood.

Shooting it as 90s found footage solved two problems at once. A found-footage treatment sits over everything and blunts the tell-tale sheen that gives AI media away. And found footage does not have to be continuous, so it can jump forward and back in time with nothing matching the shot before it. Given how inconsistent the tools were, that freed us from forcing a continuity they could not hold, and let the narrative and the feeling carry the film instead.

Sound and image drift out of sync within the first fifteen seconds. It deepens the unease, and practically, it sidesteps AI's struggle with lip-sync.

The stack

Midjourney to Runway to Premiere.

The look began in Midjourney, prompted toward a 1990s amateur, liminal, low-contrast camcorder feel. Those stills went into Runway to generate the motion. From there the footage was uprezzed, passed through the found-footage treatment, and edited in Adobe Premiere.

The constraint was the point. A flawed toolset, a small team, and 48 hours are not much on their own. What carried it was a clear creative vision for what the film should be, and the production discipline to push every limited resource to its fullest, deciding what to generate, what to keep, and what the imperfections were for.

Let's talk.

← Back to work