7-step MCP for Blender workflow lets Claude build photorealistic scenes end to end
sidahuj · x · 2026-11-03
Indie developer hayashimon used MCP for Blender to have Claude Opus 5.5 model, texture, and animate a cinematic rainy-night laundromat entirely inside Blender — driven almost entirely by Japanese instructions, no manual Blender work. The original post got 700K+ views and many followed the steps.
The 7-step layered workflow: (1) generate a 2D reference image in ChatGPT; (2) Claude reads layout and builds the room; (3) blockout — boxes first, then space, models, materials; (4) procedural grime — chipped paint, dust, drips, fingerprints; (5) printed matter — faded, peeled signs and stickers; (6) human critique, Claude checks close-ups and fixes; (7) motion — dryer spin, flickering lights, rain, camera move, rendered to video.
Key split: the human handles aesthetic judgment and change requests; Claude does all Blender operations.
More from coding & agent
- MCP for Blender hits 26.6k GitHub stars, demo builds abandoned mall with Claude — sidahuj · 2026-11-04
- Dev builds 'agent slop map' to visualize AI-generated code rot, plans to open-source — vincent_koc · 2026-11-04
- DR Pixel: open-source page-image deep research agent, 27B model matches GPT-5.4 — sewon__min · 2026-11-04
- Agentic Trading Lab: open-source Python playground for LLM-powered trading agents — mdancho84 · 2026-11-04
- Factory launches Deployment team, cites Nvidia and Blackstone as users — maddiehfaulkner · 2026-11-04
- Formally verifying RISC-V CPU mul/div instructions with one AI theorem-prover prompt — satnam6502 · 2026-11-04