No prompt one-shots a polished 3D asset — human-judgment iteration took 17 more minutes
chaseleantj · x · 2026-09-13
In his 3D generation experiment, the author argues there's only so much one-shotting can do: no prompt and no current model can produce a polished project in one shot. At some point you need human judgment.
He asked the model to combine the leaf detail and lighting of the left output with the nicer shape of the right one — a final iteration that took 17 more minutes.
More from coding & agent
- Stop harness hopping: model and endpoint decide agent success, not the wrapper — aidenclarke_12 · 2026-09-13
- Dev burns Claude's 5-hour limit in one hour, switches to Codex for implementation — rudrank · 2026-09-13
- kalomaze: enough HuggingFace slop code exists that agents now copy it by default — kalomaze · 2026-09-13
- Stripe's Jeff Weinstein: agentic purchasing via Link got visibly better in a week — jeff_weinstein · 2026-09-13
- agent-skills launches a secure, validated skill registry for AI coding agents — tech-leads-club · 2026-09-13
- Stripe's agent wallet SDK sees fast adoption months after launch, says exec — jeff_weinstein · 2026-09-13