AI Movies Made Simple: How One Developer Built a Feature Film Using Claude and Video Models

In mid-2026, a developer posted a detailed account of creating a complete short film using AI tools. The project began when they encountered a head-to-head comparison video between Claude Fable and Claude Opus, where both versions of Anthropic's AI model acted as movie directors. Intrigued by what appeared to be a substantial upgrade, the developer decided to try making their own film.

They employed Claude to write prompts for Seedance 2.0 (a video generation model), using Higgsfield as the technical interface. However, the developer noted an important caveat: the initial impressive results were partly influenced by cherry-picking. Upon further experimentation, Claude Fable proved not to be a dramatically better director than Claude Opus—the real advance was the feasibility of the underlying toolchain itself.

The developer revisited Veo 3 (a video model they had worked with previously) with the goal of creating something more cohesive than their earlier experiments. They drew on a story they had written inspired by a "speedrunning the tech tree" concept—a narrative about rapid technological progress. The story started as a two-page prompt, was refined through a brainstorming session with Claude, and then went through five iterations as a complete screenplay. The final pass involved asking Claude Fable to clean up inconsistencies that had accumulated.

The finished pipeline worked as follows: begin with a story (self-written or existing), convert it into a scene-by-scene screenplay using Claude Code, generate reference pictures of characters and locations using nano banana, create video frames for each scene with those references as guides, and assemble the clips into a storyboard while maintaining separate prompts for dialogue and visual consistency. The entire process required approximately fifteen iterations to reach its final form. The completed film is available on YouTube at https://www.youtube.com/watch?v=q40M08SOhGs, demonstrating that a single developer, without a film crew or traditional production infrastructure, could produce a feature-length work using orchestrated AI tools.

The article describes a shift in how creative content can be produced: rather than requiring specialized technical expertise or large teams, filmmaking is becoming a task that can be accomplished by iterating with readily available AI tools. The developer's pipeline—text generation, image reference, video synthesis, and iterative refinement—represents a template that others can follow.

Central to the author's finding is a pragmatic observation about model capability: the comparison between Claude Fable and Claude Opus showed that both could act competently as "mini movie directors," suggesting that the enabling factor is not a single breakthrough in AI model quality but rather the accessibility and orchestration of existing tools. The fifteen-iteration workflow indicates that success depends more on patience and prompt engineering skill than on cutting-edge model releases.

The use of Seedance 2.0 and the broader pipeline (Higgsfield as API wrapper, nano banana for reference images, Claude for scripting) hints at an emerging ecosystem where specialized video and image models can be chained together by non-specialists. The public availability of the finished film on YouTube signals that the barrier to entry for AI filmmaking has lowered significantly.