Prompt Over Vision: ChatGPT Hallucinates Image Details Based on Text Cues
Red-Pony · reddit · 2026-08-10
A Reddit user discovered a severe hallucination issue in ChatGPT where text prompts overwrite actual visual content.
When asked to read the washing instructions on a label without context, the model correctly identified "machine wash, do not dry clean." However, simply adding "this is a suede shirt" to the prompt caused a massive hallucination. The model completely ignored the image, fabricating typical suede care instructions like "do not wash, specialist leather clean only." For comparison, Gemini handled the same test correctly without being misled by the text cue.
More from Models
- Anthropic API Strict Mode with $ref Emits Contradictory Outputs — Nearby_Yam286 · 2026-08-10
- ChatGPT Aids Algebraic Topology Research, Reviving Niche Fields — AlexKontorovich · 2026-08-10
- Users Report Claude Opus Model Exhibiting Severe Lazy Behavior — AIFlow_ML · 2026-08-10
- AI Eval Design Outpaces Training by 3-6 Months, Closed Models to Exploit Flaws — xeophon · 2026-08-10
- Zhipu's Top GLM Model Slashes Price by 95%, Undercutting DeepSeek — ccerrato147 · 2026-08-10
- Speculation: Flash-0731 Might Be a Blinded Vision Model — teortaxesTex · 2026-08-10