Agentic 3D rendering: new LLM APIs lose to training-distribution-familiar interfaces, with DeepSeek demo
teortaxesTex · x · 2026-11-08
The author, building an agentic rendering pipeline, reports his preliminary conclusion: designing new APIs for LLMs isn't worth it. For frozen-weight models, the salient measure of 'optimality' is similarity to the training distribution — and your new API isn't there. He installed vork as a DSH-plugin, finds it neat but a small slice of what's needed, and says Blender MCP remains indispensable. Screenshots show DeepSeek-V4.1-Flash planning a high-quality mushroom render over 130 steps (99% cache hit), working around missing primitives by scaling spheres and using std/sky environment materials.
Related event: AI-Oriented APIs Aren't Worth It; Training Distribution Matters More(2 posts)→
More from coding & agent
- DIY LLM vulnerability scanner finds 10 bugs for ~$20/day, 8 novel vs CodeQL — _h4xr · 2026-11-08
- leaf: a terminal Markdown previewer with GUI-like rendering and Mermaid support — 4310sy · 2026-11-08
- Perplexity CEO on Joe Rogan: the desk job was a sales strategy, and AI taking it over isn't bad — 2C_ornot2C · 2026-11-08
- blank: open-source local-first agent harness fixes parallel sub-agent collisions and rate-limit halts — Big_Veterinarian7479 · 2026-11-08
- Alibaba's Co-Design Bench: top model fully solves only 26.4% of real coding tasks — The Decoder · 2026-11-08
- Readable Node: an agent skill that rebuilds messy ComfyUI workflows into readable ones — nomadoor · 2026-11-08