Distinguished Scientist at Google DeepMind. Gemini Multimodal Lead. My own views here.
Los Angeles, CA
Joined June 2018
- Tweets24
- Following71
- Followers315
- Likes109
Meet Vision Banana 🍌 from @GoogleDeepMind! We provide strong evidence that image generators are generalist vision learners. Traditional computer vision tasks (segmentation, depth estimation, normal prediction) can now be performed at/near SOTA with a single generalist model derived from an image generation model.
🖼️ Explore the results: vision-banana.github.io
📄 See details at: arxiv.org/abs/2604.20329
Radu Soricut retweeted
Gemini 3 Pro is the frontier of multimodal AI, delivering SOTA performance across document, screen, spatial, and video understanding.
Read our deep dive on how we’ve pushed our core capabilities to power hero use cases across:
+ Docs: "derender" complex docs into structured code (HTML/LaTeX)
+ Screen: build robust computer agents that automate complex tasks
+ Spatial: generate collision-free trajectories for robotics & XR
+ Video: analyze sports footage using high-FPS processing with "thinking" mode
See how these capabilities are transforming workflows in education, biomedical, and law/finance → goo.gle/3Mt3UlT
Radu Soricut retweeted
I got a bit emotional seeing this illustration from Nano Banana Pro, because it depicts one of our major discoveries from more than 15 years ago, & it is a beautiful & accurate illustration; NB Pro knew it all. 🥹
Prompt: “Draw an illustration of GARP biology for Tregs”
Radu Soricut retweeted
We're doing our best to keep the chips running 🚒
Banana command center activated! Hey hey @OriolVinyalsML rumor has it you showed up to work in a full banana uniform today. Can you confirm or deny?
With Nano Banana Pro (deepmind.google/models/gemin…), we have come such a long way in such a short amount of time. In 9 months we’ve launched 3 models, each iteration making the previous ones pale in comparison. I cannot be prouder of this team and what we delivered!
This is not just about the English-speaking world. Here’s a cool one (issued in the same context as the one about the US presidents): “Do the same for the Spanish wars of the 20th century, using Spanish to render the text.”
Note how amazing it is that Franco actually gets older!
1
72
Learning visually with GemPix also applies to Humanities. Here’s an example of how you can encapsulate an interesting topic about US History issuing a simple prompt like “Show me in an infographic the American presidents that led the US through significant wars”.
1
1
68
If you are a visual learner, what Nano Banana Pro does is a game changer. Consider this rendering below starting from the simple prompt “Visualize Bernoulli's theorem”. For most students who look at the equation alone and wonder what it means, this can be an Eureka moment:
3
162
It’s out into the world for Gemini 2.5 Flash Image (aka nano-banana) after an amazing sprint that got us from the previous model to this one. I couldn’t be prouder of the whole Google DeepMind team that made this happen.
2
14
581
The level of consistency of this model when Editing is unprecedented. The LMArena surely took notice.
1
67
The 3rd 🚀 today, the one we sweated a lot over -- Native image generation with Gemini 2.0 Flash available to all developers as an experimental release in the Gemini API and Google AI Studio:
developers.googleblog.com/en…
