Users report GPT gradually corrupting detailed image prompts across turns
apostrophefee · reddit · 2026-09-25
A user complains that GPT gradually rewrites their detailed H3 image prompt across a conversation, adding and corrupting details along the way. They need an LLM to handle long, precise prompts and find the experience frustrating — contrasting with the popular belief that AI just magically produces whatever you want.
More from Fun
- Turning an AI-generated image into a playable game in 5 minutes — gizakdag · 2026-09-25
- NeurIPS results day exposes peer review double standards, researchers joke — prajdabre · 2026-09-25
- Codex skips its web search tool, opens Google in browser instead — TheIshanGoswami · 2026-09-25
- Muse invite codes resold for 29.9 yuan on Xianyu despite free signup — sven_ai · 2026-09-25
- Self-play RL finally cracks Brood War — with a 300M-parameter model — syhw · 2026-09-25
- Have You Tried Kicking the Rack Over to Reseat the GPUs? — sloppenheimer · 2026-09-25