原帖内容

GPT-Image 2.5 just solved character consistency. Here's the back-to-back comparison along older models: