新版GPT Image 2.5已经能伪造GPT-6发布会了

TL;DR - Reports of an unreleased “GPT Image 2.5,” possibly appearing anonymously as luna-lisa-alpha on LMArena, suggest major gains in photorealism, text rendering, speed, and character consistency. Its ability to create convincing fake event and social-media images also highlights growing misinformation risks.

• Some ChatGPT users reportedly saw a pop-up naming GPT Image 2.5, but no official release is cited.
• Test images reportedly show fewer noise artifacts and more realistic scenes than GPT Image 2.
• The model appears better at preserving faces across multiple images and rendering embedded text without distortion.
• Examples include fabricated GPT-6 presentation scenes and convincing screenshots attributed to public figures.