@teortaxesTex: My personal simple trial for t2i models: feed a real image to a good captioner (eg 3.5-Flash) to generate a prompt, then have it be executed...

@teortaxesTex 3 信息等级 3 发布：2026-06-09T22:07 抓取：2026-06-09 23:18

🔗 原文链接

AI 行业

摘要

用户通过图像描述器生成提示词，测试了HiDream-O1-Image和GPT-Image-2的文本生成图像能力，发现HiDream-O1-Image在逼真度上优化强势，但不及GPT-Image-2。

客观事实

用户用真实图像生成prompt，测试t2i模型
HiDream-O1-Image在photorealism上优化强
HiDream表现不如GPT-Image-2

GPT-Image-2 HiDream-O1-Image

原文

My personal simple trial for t2i models: feed a real image to a good captioner (eg 3.5-Flash) to generate a prompt, then have it be executed.
The HiDream-O1-Image is strongly optimized for photorealism. But it's not really in GPT's tier.
Original photo, GPT-Image-2, Hi-Dream. https://t.co/WWTd5rjyKc

likes: 4 | retweets: 0 | replies: 0 | views: 955