Summary
This hands-on review puts Anthropic’s newly released Claude Fable 5.2 through a series of demanding creative and agentic tests, mostly run through the Claude Code harness so the model can work with local files and folders across projects.
The first test asks the model to build a 90-second, 1080p explainer video on the history of humanity through 2030 using pure JavaScript, with a voiceover generated via Gemini TTS. The model planned the work, inspected its own frames, fixed errors autonomously, and even added background music unprompted after a revision pass that brought the total time to roughly an hour and a half. Further demos include operating a virtual piano in a browser, generating a 3D architectural render of a villa that the presenter calls the best result so far among frontier models, and vibe coding a third-person ninja-style game using Blender, Unreal Engine, a Sketchfab character, and Adobe Mixamo animations.
The video also shows how connecting Higgsfield to Claude can add image and video generation through models like Seedance, Kling, GPT Image, and Nano Banana. Viewers get a practical look at what Fable 5.2 can do on long-running, multi-tool tasks, along with usage stats for each session.
📺 Source: AI Search · Published November 11, 2026
🏷️ Format: Review
