Top
Best
New

Posted by davidbarker 11 hours ago

Nano Banana 2: Google's latest AI image generation model(blog.google)
487 points | 470 commentspage 4
rosstex 9 hours ago|
Adding to predictions: the magic of travel might actually be reborn, as people seek authentic experiences.
jslakro 9 hours ago||
It'll be great to find a web directory dedicated exclusively to good/useful prompts with nano banana
hmokiguess 10 hours ago||
The Chinese are so much ahead in this space, their models are way better at this stuff. For example, https://hunyuan.tencent.com/image/en?tabIndex=0 and https://seed.bytedance.com/en/seedream5_0_lite
sigmar 10 hours ago||
which models?

we have user-preference rankings that put NB2 on top: https://arena.ai/leaderboard/text-to-image

hmokiguess 10 hours ago||
interesting, maybe this is just anedoctal experience then and I'm biased but I have been preferring theirs over Nano Banana
raincole 9 hours ago|||
When it comes to image prompts, Seedream is far behind Nano Banana. "Far behind" is a ridiculous understatement here, btw.

Afaik the only real competitor is Riverflow V2.

KK7NIL 10 hours ago||
Kind of a pointless comment without a link to such a model.
hmokiguess 10 hours ago||
sorry, I've edited my original comment with one such example, but they're easily discoverable I assumed it was popular knowledge

- https://hunyuan.tencent.com/image/en?tabIndex=0

- https://seed.bytedance.com/en/seedream5_0_lite

someone shared benchmarks that differ my experience tho, so I may be biased

sync 11 hours ago||
Did gemini-2.5-flash-image get an upgrade as well? I just got the following, which is fascinating, and not something I've seen before:

> I'm sorry, but I cannot fulfill your request as it contains conflicting instructions. You asked me to include the self-carved markings on the character's right wrist and to show him clutching his electromancy focus, but you also explicitly stated, "Do NOT include any props, weapons, or objects in the character's hands - hands should be empty." This contradiction prevents me from generating the image as requested.

My prompts are automated (e.g. I'm not writing them) and definitely have contained conflicting instructions in the past.

A quick google search on that error doesn't reveal anything either

JOJESU 10 hours ago||
I’ve been exploring this exact problem space from the angle of extreme constraints (single-digit MB memory, no cloud assumptions). I documented what broke first and why here, in case it’s useful: https://github.com/nullclaw/nullclaw
nathan_compton 10 hours ago|
So this is an ultra-minimalist software platform to farm work out to enormous energy chugging AI models?
evrenesat 10 hours ago||
I only needed help of this banana boy twice, it managed to disappoint me each time. The most recent one, I was trying different beard and mustache styles on myself, on a photo I imported from my own Google photo gallery, and it consistently rejected me, claiming I'm a public figure. Nobody ever told me that I look like any famous person, so that's googles own bananination. ChatGPT nicely handled the job.
dgtlanml2 11 hours ago||
Wow the article narration with Umbriel is silent after the 6 second mark.
ozgung 9 hours ago||
Any info or speculation about technical details?
mowmiatlas 9 hours ago||
So, I'd suspect the seedance2.0 competitive video model is coming as well soon? ;)
sorenjan 11 hours ago|
Is this a distillation of Nano Banana Pro?
meetpateltech 11 hours ago|
Gemini 3.1 Flash Image is based on Gemini 3 Flash.

source: https://deepmind.google/models/model-cards/gemini-3-1-flash-...

More comments...