Quick Overview
In this demonstration video, creator Nate Herk evaluates the coding capabilities of Anthropic's Claude Code environment. He inputs an online prompt into Opus 5.5 to determine whether an AI coding agent can autonomously generate a complete 3D animated scene in Blender from scratch.
Key Points
- 1.A 390-word prompt provided to Opus 5.5 within Claude Code generated an entire 3D animation workflow from code.
- 2.The model generated 12 files that covered the interface, the 3D Blender scene, the soundtrack, and the rendering pipeline.
- 3.Initial test renders contained visual defects, including misplaced buttons, off-screen title text, and an incorrect salmon color on the character.
- 4.Claude reviewed 18 half-resolution still frames to identify mistakes, corrected the script, and re-rendered the scene to fix the issues.
- 5.The complete process rendered a 12-second, 30 frames per second motion graphic in 1080p resolution.
Summary
Nate Herk demonstrates an automated 3D animation workflow created entirely through code using Claude Code running Opus 5.5. The experiment begins with a copy-ready 390-word prompt taken from an online source, targeting a 12-second, 30 frames per second motion graphics video at 1920x1080 resolution set to 120 beats per minute. The prompt defines Clawd, the Claude Code mascot, moving across a pixel-faithful After Effects workspace, with instructions to execute the render using Blender Python and Cycles GPU.
Upon receiving the prompt, Claude Code generates a project structure consisting of 12 separate files. These include timeline.py, build_ui.py, render_ui.js, HTML source files, scene.py for Blender scene setup, gpu.py, render.py, render_chunks.sh, sound.py for synthetic audio generation, and encode.py. The Python scripts programmatically set keyframes for the camera, the root object, and UI elements, while sound.py generates a lo-fi beat and sound effects directly in code.
The initial render preview uncovers multiple compositional errors. Reviewing the contact sheet of 18 half-resolution test stills reveals that interface buttons and the playhead landed in incorrect coordinates, the title text ran off the right edge of the frame, and the mascot appeared salmon-colored rather than the intended shade. Claude analyzes these still images, detects each discrepancy, modifies the offending code files, and re-renders the assets. The resulting final render produces a fully corrected 3D motion graphic with aligned interface elements, accurate colors, and audio.
Prompting Claude Code to Generate 3D Assets
Nate Herk tests an online 390-word prompt using Opus 5.5 in Claude Code to see if the model can construct a complete 3D animated scene. The prompt specifies a 12-second animation featuring the Claude Code mascot Clawd inside an After Effects-style workspace, rendered in Blender Cycles GPU.
Multifile Script Generation and Initial Renders
Claude Code writes 12 distinct files handling the user interface layout, the Blender 3D scene, Python-generated sound, and the full rendering pipeline. The preliminary render contact sheet reveals visual errors, including misplaced interface controls, text clipping off screen, and an inaccurate pinkish color value for the character.
Visual Inspection and Autonomous Error Correction
Claude inspects 18 rendered test stills, logs the discrepancies between the expected values and rendered frames, adjusts the underlying code, and re-executes the render pipeline. The final output produces the corrected motion graphic with accurate placement, color matching, and synchronized audio.
The Bottom Line
The video establishes that Opus 5.5 within Claude Code can translate a multi-part text specification into a working 12-file codebase that produces a complete 3D animation in Blender. It demonstrates how autonomous evaluation of rendered still frames enables the model to identify visual mistakes and adjust its own code without manual keyframing. The presentation concludes by displaying the finished animation and offering the prompt guide, leaving broader production scalability across longer or more complex scenes unexamined.
FAQ
What is Opus 5.5 and how does Claude Code use it to create 3D renders?
Opus 5.5 is an AI model used inside Claude Code to programmatically write and execute scripts for Blender, producing a complete 3D animation, interface graphics, and sound from a text prompt.
What files were generated by Claude Code to build the Blender 3D animation?
Claude Code generated 12 files, including scene.py for the Blender scene, sound.py for the soundtrack, build_ui.py and HTML templates for interface assets, and various rendering scripts like gpu.py, render.py, and encode.py.
What errors were identified in the first test frames of the 3D animation?
The initial preview showed buttons and the playhead in the wrong spots, the character mascot colored pinkish or salmon instead of the target hex code, and title text clipping off the right edge of the screen.
How did Claude Code correct the rendering errors found in the initial output?
Claude reviewed a contact sheet of 18 still frames, logged the coordinate and color discrepancies against the prompt requirements, updated the scripts, and re-rendered the scene with the fixes applied.
What are the format specifications for the 3D animation generated with Claude Code and Blender?
The target format specified in the prompt is a 12.0-second video at 30 frames per second in 1920x1080 resolution, timed to 120 BPM and rendered using Blender Cycles on GPU.
Worth watching for
Developers, technical artists, and 3D animators interested in programmatic 3D scene generation and automated multi-file code pipelines using LLMs and Blender.
- claude-code
- blender
- opus-5-5
- python
- 3d-animation
- motion-graphics