Prof's trick: restyle a screenshot in ChatGPT, feed it to Codex as goal image
CSProfKGD · x · 2026-10-03
Georgia Tech professor K. Gopalakrishnan (CSProfKGD) shared how he iterated on UI layouts with his wife. The hardest part was getting Codex to nail a clay-like visual style from text prompts alone.
His solution is a two-step workflow: drop a screenshot into ChatGPT and ask it to restyle the UI as clay, then use the result as the goal image in Codex. Codex then autonomously ground away for about half an hour and came very close to the target. The trick—generate a style reference image first, then hand it to a coding agent—is directly reusable for UI work.
More from coding & agent
- Jev open-sourced: PowerShell module turns natural language into typed AI decisions — dfinke · 2026-10-03
- Video explains Jev agent examples: games, triage, and browser use — RelevantEmergency707 · 2026-10-03
- Making Claude music videos: upload your own audio instead of browser-driving Suno — technollama · 2026-10-03
- 'AI Babysitter' Is the New Job: Keeping Agent Queues at 100% Utilization — petesena · 2026-10-03
- Antigravity CLI 1.2.15 runs the full coding agent on Android via Termux — kalyan_kpl · 2026-10-03
- OpenAI DevDay: Each Dot Agent Gets a Dedicated Cloud Computer on GPT-6 Astra — dl_weekly · 2026-10-03