ChatGPT Plus users report image generation quality appears degraded
nilanganray · reddit · 2026-09-16
Reddit user nilanganray reports that ChatGPT image generation has visibly declined over the past 3-4 days: sub-standard outputs despite good prompting, alignment/clipping/centering issues unseen in three months, and revision prompts misunderstanding context and changing unrelated elements — on Plus plan Work mode, with chat mode oddly slightly better.
The post also makes a broader point: even with identical prompts and workflows, AI products can't be relied on long-term — LLMs already lack predictability, and labs change things on a whim.
More from Models
- DeepSeek V4.1 Flash ignored by eval community despite its significance — teortaxesTex · 2026-09-16
- Community squeezes a 124B model onto a 128GB DGX Spark with quantization and kernel fixes — alifcoder · 2026-09-16
- Dev: Claude, GPT and Gemini all seem to go dumb on the same day — dejanseo · 2026-09-16
- What's new in Gemini Live API: new audio models explained — patloeber · 2026-09-16
- Coding-specialized DeepSeek V4 Flash hurts long-context novel and RP users — teortaxesTex · 2026-09-16
- Codex power users track daily limit resets with a 3-6 hour warning pattern — ___Patrice___ · 2026-09-16