Autodesk Made an AI Film Then Found a Mistake It Couldn’t Fix

One of the world's biggest makers of professional film and animation software has created an AI-assisted movie of its own, only to discover a mistake that even its team couldn't persuade the technology to fix.

Autodesk is the US software company behind tools used throughout the film, television, animation and visual effects industries. Its products include Maya, a 3D animation program that has been used by professional studios to create characters, environments and visual effects.

Now Autodesk has been experimenting with how generative AI could fit into filmmaking by creating Robots of the Wild West, a demonstration film following robots Rusty and Lasso Liz as they try to save their town from a gold miner.

But the experiment also exposed one of AI video's remaining problems. During one scene, a character is sitting next to another character at a table. When the camera changes angle a couple of shots later, he has mysteriously moved and is now sitting across the corner from her.

Nobody moved him as part of the story. The AI simply failed to remember exactly where he was supposed to be. Filmmakers call this a continuity error, which simply means something unexpectedly changes between shots even though it should have remained the same.

Autodesk spotted the problem and tried to fix it. Maurice Patel, the company's vice president of media and entertainment industry strategy, explained that the team changed its AI prompts, the written instructions telling an AI system what it should generate, but couldn't produce the result it wanted.

Rebuilding the sequence would have required more work than Autodesk considered worthwhile for a demonstration, so the mistake stayed.

The company is also developing Flow Studio, filmmaking software designed to give creators more control when using AI. Rather than producing one finished video that is difficult to change, it can combine AI-generated material with editable 3D elements such as characters, cameras and environments.

The experiment provides a simple example of why AI filmmaking still needs human involvement. Generating an impressive individual clip is one thing, but keeping every character, object and position consistent across a complete scene can be considerably harder.