For $10, two hours, and 5,500 lines of code, Kapasi used Claude Opus 5 to build the 3D world of "Lord of the Rings"

Zhidongzhi Compilation | Eggplant Editor | Cheng Qian Zhidongzhi reported on August 3 that yesterday, OpenAI co-founder, current Anthropic pre-training team member, and AI guru Andrej Karpathy demonstrated a new method of model testing at X. He asked Claude Opus 5 to use Three.js to make the opening story of "The Lord of the Rings" into a 3D animation with a budget of 1 million tokens. In the end, Claude Opus 5 ran for about 2 hours and wrote about 5,500 lines of code. The task cost was about US$10 (approximately RMB 68). Kapaci thought that although the finished film was a bit rough, it was interesting. ▲Kapasi’s tweet (Source: Claude Opus 5 requires understanding the original text, disassembling the scene, creating polygonal assets, placing characters and objects into three-dimensional coordinates, writing lens and animation codes, and checking and modifying the results during operation. The final deliverable is not a picture synthesized by the video model, but a Three.js program that can run in the browser. ▲Claude Three.js animation screen generated by Opus 5 based on the opening chapter of "The Lord of the Rings" (Source: X) However, the test also exposed the shortcomings of the model. Claude Opus 5 cannot efficiently watch continuous videos, nor can it enter scene trials. It can only slowly intercept pictures at different points in time and check them one by one. Capasi said that Claude Opus 5 made several mistakes as a result, and the final film still had many rough spots. Roope Rainisto, creative AI director of Finnish mobile game company Supercell and AIGC creator, proposed inputting programmed animation into video models such as Seedance to improve picture quality. Kapasi agrees with this idea and believes that the procedural code can be responsible for storyboarding and control, while the video-to-video model is responsible for texture.