The line between what's AI and what's real just got even more blurry...
One shot image editing for realistic facial expressions is solved, thanks to the unreleased model in LMArena "nano banana"
A comparison of the previous best image editing model (Kontext) vs this new SOTA model
The difference is crazy 👇
The most obvious improvement is prompting for "happy". The smile appears way more natural and less plastic. https://t.co/pbS1lodtaq
Kontext tends to over-exaggerate expressions, making the subject look like a caricature of themselves. https://t.co/UFboWi1eW1
Kontext really went for it, but still too much. https://t.co/TLbABrl1mk
For winking, nano-banana handled it with ease while Kontext seems to not understand the concept. https://t.co/HFaqJCEvNC
Another one on the scoreboard for nano. https://t.co/fNOYLfjkGq
These could possibly go either way, but I'd still give the edge to nano. https://t.co/fDbHICJWTt
