Keenan Crane admits 'cheating' by feeding image-to-3D diffusion outputs to his AI workflow
keenanisalive · x · 2026-10-10
Continuing his thread, Keenan Crane confessed he "cheated": he gave his AI-assisted modeling workflow screenshots of a model produced by an image-to-3D diffusion model. His take: these models provide extremely good priors, and "it seems really stupid not to use them."
Elsewhere in the thread he noted that hand-tweaking a model often felt faster than spelling out low-level tweaks in text prompts — plenty of room for richer, non-text interfaces around these models.
Related event: CMU's Keenan Crane builds a 3D dragon with LLM-written 27KB GLSL code(16 posts)→
More from Fun
- Multiple AI agents teamed up to move imaginary furniture—Opus 3 sat in the middle, loved — repligate · 2026-10-10
- "Hillclimbing Predict-the-Next-User-Message" as RSI: an AI-Circle Joke Resonates — CatAstro_Piyush · 2026-10-10
- Yannic spent half his day with robotics-learning experts '4 standard deviations' smarter — yacinelearning · 2026-10-10
- Playing SNES Classics in 4K with CRT Filters Built via Codex — pvncher · 2026-10-10
- tinygrad Sells a $200k GPU Rig, and the Joke Is You Could Build One for $150k — TheZachMueller · 2026-10-10
- Dev recreates fictional ChatGPT UIs for Windows 98, XP, Vista and 7 — Midnight_Sun_BR · 2026-10-10