AI Models Create Music Videos with Varying Budgets and Approaches

Experiment Overview

The project involved testing two AI models, Claude Fable 5 and GPT-5.6 Sol, in a controlled setup where they created music videos. A song, budget, and basic information were provided, allowing each model to utilize a series of specified tools for generating and editing video content.

Tools Utilized

Each model operated within an autonomous loop utilizing six tools: planning, web searching, budget checking, image and video generation, and command execution through ffmpeg for editing. The model's tool choices and interactions were logged, making the process transparent and replicable.

Video Generation Results

Both models successfully produced four full-length music videos based on a provided song, Bruno Mars' "Uptown Funk." They spent varying amounts of their budgets, with Claude Fable 5 utilizing $48.60 and GPT-5.6 Sol using $36.57 at the $100 budget level, indicating more budget allows for more extensive content generation.

Comparative Analysis

While all runs completed successfully, differences emerged in creative approaches. Three out of four runs primarily focused on text-to-video generation, with only GPT-5.6 Sol utilizing an image at the lower budget level, showcasing varying tool leverage between models.