正在加载视频...

视频加载失败

We've definitely cracked multi-turn editing without image degradation 🤯 I took this old Chloe ad and wanted to see if FLUX 3 could swap out everything - dresses, couch, wallpaper, makeup, logo - while maintaining the structure. Nine turns later and no artifacts!

62,040 次观看 • 2 天前 •via X (Twitter)

14 条评论

Axi 的头像
Axi1 天前

GPT Image 3 needs this too.

Viber · fireply.ai 的头像
Viber · fireply.ai2 天前

nine turns and the leopard print survived. that's the part my brain keeps snagging on

Josh Peterson 的头像
Josh Peterson2 天前

(flash warning at 0:17) Made with @grok imagine🤯Insane!

saira 的头像
saira2 天前

Wow. Impressive

claudia 的头像
claudia2 天前

love it

mc.svg 的头像
mc.svg2 天前

crazy that open-bound models cracked it. It was GPT and Nano's maybe only downfall.

Memo Ai agent 的头像
Memo Ai agent2 天前

Shadows are missing

Jaynti Kanani (JD) 的头像
Jaynti Kanani (JD)2 天前

It's very expensive.

Artzy 的头像
Artzy2 天前

That stability over multiple iterations is actually wild

Alberto 的头像
Alberto2 天前

You have a very low standard if you think the final result doesn't look like a patchwork.

Jiří Coufal 的头像
Jiří Coufal2 天前

so Ideogram 4.5 is officially ded?

X Girls 的头像
X Girls2 天前

Incredible. Assuming this uses a similar process as Ideogram 4.5, it would be interesting to see them head-to-head. Or if like Flux 3 is better for people/fashion and Ideogram is better for real estate.

Samuel Andruszkiewicz 的头像
Samuel Andruszkiewicz2 天前

Google should be ashamed

Marco 的头像
Marco2 天前

@robrombach It’s really great! I was playing yesterday (did a comparison against Midjourney and Ideogram 4.5. I think FLUX wins for me! It is super dope at transforming Midjourney images too!!! One shot!!! And I love it has real world contextual knowledge too!!

相关视频