Project: Baha'i Temple of South America
Archtecture: Hariri Pontarini Architects
Photo: Doublespace Photography
Geo: Santiago, CL
Hello everyone. Don’t blame me for the silence on air: I’m scolding myself for the fact that my body sometimes demands time to sleep. First of all, it’s the end of the year; secondly, every week there’s some kind of AI revolution that changes workflows and requires quick implementations. There are special resources to track these changes, so I usually focus on something key. We’re more about architecture here.
This time, the subject is the update of the nanobanana to version 3 Pro. There are two key changes: brains from Gemini 3 and native 4K. It’s important to note that I couldn’t test 4K: the output is in the classic 1024 resolution both in Google’s built-in interface and in the Photoshop license, which received the banana update today. Online, people say that 4K was enabled briefly, but due to long generation times, it was decided to revert to the usual 1024.
You can read a bit about what a banana with Gemini brains is in the official release: deepmind.google/models/gemini-image/pro/
The release evoked such emotions in me that I disturbed the whole team before bed and philosophically discussed with those on the other side of the globe at 2 a.m., asking what comes next?
There are many thoughts, and I won’t calm you down. This goes beyond just a simple pipeline tool, as tasks are being solved that previously could only be done by skilled specialists: people, seasons, atmosphere, details, content, plants, and everything else your heart desires, natively out of the box, using not-so-smart prompts.
But all of this pales in comparison to what I saw from Sergey Tsypkin, where he uploaded house drawings in all projections into the banana and got a render out in the same projections while preserving almost all details. I briefly repeated this trick on one of the old client projects: not everything is perfect, of course, but it works. I uploaded facades, a general plan, angle descriptions, specifications, facade drawings, etc., and got a render in return! It will do for a conditional sketch project stage, but that’s it for now.
The difficulties in providing access to 4K indicate that even giants like Google need hardware and a lot of it. You can’t just snap your fingers and get it, so at least in this direction, there’s a slight advantage: use it wisely.
Expertise, visual experience, and the ability to solve problems on the fly rather than on paper—this will be valued, but entering this circle by just sitting in front of a window with GPT won’t work. LLMs are a set of weights, and for them to work correctly for you is also an art.
PS A question for experts: Can billing be linked to the Gemini API from Russia? I’ve tried everything (various foreign cards, clean U.S. Google accounts, different VPNs—nothing works).
@gorkjournal
3D · Render · Photo · Archviz · GORK
Archtecture: Hariri Pontarini Architects
Photo: Doublespace Photography
Geo: Santiago, CL
Hello everyone. Don’t blame me for the silence on air: I’m scolding myself for the fact that my body sometimes demands time to sleep. First of all, it’s the end of the year; secondly, every week there’s some kind of AI revolution that changes workflows and requires quick implementations. There are special resources to track these changes, so I usually focus on something key. We’re more about architecture here.
This time, the subject is the update of the nanobanana to version 3 Pro. There are two key changes: brains from Gemini 3 and native 4K. It’s important to note that I couldn’t test 4K: the output is in the classic 1024 resolution both in Google’s built-in interface and in the Photoshop license, which received the banana update today. Online, people say that 4K was enabled briefly, but due to long generation times, it was decided to revert to the usual 1024.
You can read a bit about what a banana with Gemini brains is in the official release: deepmind.google/models/gemini-image/pro/
The release evoked such emotions in me that I disturbed the whole team before bed and philosophically discussed with those on the other side of the globe at 2 a.m., asking what comes next?
There are many thoughts, and I won’t calm you down. This goes beyond just a simple pipeline tool, as tasks are being solved that previously could only be done by skilled specialists: people, seasons, atmosphere, details, content, plants, and everything else your heart desires, natively out of the box, using not-so-smart prompts.
But all of this pales in comparison to what I saw from Sergey Tsypkin, where he uploaded house drawings in all projections into the banana and got a render out in the same projections while preserving almost all details. I briefly repeated this trick on one of the old client projects: not everything is perfect, of course, but it works. I uploaded facades, a general plan, angle descriptions, specifications, facade drawings, etc., and got a render in return! It will do for a conditional sketch project stage, but that’s it for now.
The difficulties in providing access to 4K indicate that even giants like Google need hardware and a lot of it. You can’t just snap your fingers and get it, so at least in this direction, there’s a slight advantage: use it wisely.
Expertise, visual experience, and the ability to solve problems on the fly rather than on paper—this will be valued, but entering this circle by just sitting in front of a window with GPT won’t work. LLMs are a set of weights, and for them to work correctly for you is also an art.
PS A question for experts: Can billing be linked to the Gemini API from Russia? I’ve tried everything (various foreign cards, clean U.S. Google accounts, different VPNs—nothing works).
@gorkjournal
3D · Render · Photo · Archviz · GORK