Video · Riley Brown ·
I Spent $5,000 Using Claude Fable 5.2 (This Feels Different)
In short
The video tests Claude Fable 5.2 across 3D production, video editing, games, app building, research, and hardware design. It argues that the model’s ability to coordinate visual workflows and tools makes it a major productivity shift for people building with AI.
Key points
- Claude generated a 4K Blender video from one prompt using Python, rendering 1,329 frames in 52 minutes on 10 RunPod GPUs for about $6.
- It edited a vertical 4K video with motion effects and sound effects from a single instruction.
- A playable game with original models, graphics, physics, and audio reportedly cost $550 in API usage.
- The creator says clear prompts let Claude build and test functional apps with near-100% success.
- Claude produced research presentations with sourced data, charts, and forecasts after analyzing reference material.
- For hardware design, Claude coordinated KiCad PCB work, Autodesk Inventor casing, and Blender rendering.
Written by an AI model from the video’s captions and description.
I spent the past 48 hours putting Claude Fable 5.2 to work across Blender, video editing, games, apps, research, and hardware. Here are the demos that made me think we've reached what I call Artificial General Visual Intelligence (AGVI).
In this story
A tool or model opens in the Heidelberg AI Curriculum: what it does, what it costs, how to start. A repo opens on GitHub.


