The new AI model by google is currently taking the internet by storm and with good reason. Tt can generate full 3D building models from 2D plans with remarkable likeness all through just text and a reference or two. Even take simple images such as google earth screen shots to create clean isometric views which can be transformed into actual 3D models.
A real competitor to flux context max and chapt gpt And what separates it out from other models is that is is free. I will be focusing on how to create clean 3d diagrams, visuals and models.
00:00 Intro
00:40 AIStudio use
01:33 Google earth to 3D
03:06 Plan to 3D image
05:12 3D diagrams, models and Isometric images
06:33 Architectural drawings
08:18 Outro
https://aistudio.google.com/
elementary school designed by Daaz Office: https://worldarchitecture.org/architecture-news/emefe/circular-wall-featuring-inhabitable-holes-encapsulates-elementary-school-designed-by-daaz-office.html
DARP built botanical garden: https://worldarchitecture.org/architecture-news/enzhh/darp-built-botanical-garden-in-colombia-with-self-supporting-glazed-volumes.html
#google #aistudio #imageto3d #architecture #nanobanana
Nano Banana, the new AI model by Google, is currently taking the internet by storm, and with good reason. It can generate 3D models of 2D plans with remarkable likeness, all through just text and a reference. It can even take simple images such as Google Earth screenshots to create clean isometric views, which can then be transformed into actual 3D models. This is a real competitor to Flux Context Max and Chat GPT. And what separates it out from other models is that it’s currently free to try out. I’ll be focusing on how to create clean 3D diagrams, visuals, and models. In this video, get started with this. Go over to AI Studio. Sign into it or make a Gmail account first. And there are a few terms that you need to agree to. On the top right, make sure that Gemini 2.5 Flash image preview is selected. This is the official name for the nano banana model. So make sure that you use this and then you are ready to go. This is also multimodel meaning you don’t need to use just text but you can upload all sorts of other media too. There is a token count at the top right which shows the limit for the current conversation. So the text and references you use have a token count. As you can see here, this text sentence has 16 tokens. If you reach the limit, just simply start a new conversation. Let’s get started by taking a reference image from Google Earth to create a 3D model. Best to set the base map to clean to remove all the labels. I will choose this block in Barcelona as it is very detailed and complex. I’m using the snipping tool just to grab a screenshot of a good angle. Then I go over back to Nano Banana and I’ll paste this image and enter. Use this provided image of an architectural block in Barcelona as a reference to create a high quality 3D background. Then add some details and then write show a view from an isometric with a white background. And then within about 10 seconds I have this very similar block but removed from a messy context. Even has the trees and dark shadows removed. And then all the other missing details have been very accurately recreated. Really quite incredible. And with such a clean isometric model, you can easily take this into a AI 3D generation model. Here I’m using Hanuan 2.5 as it creates excellent results from one image. And after 1 minute, we have the Google Earth block fully recreated with all the details. Even the complex roof levels have come in. only the interior court to yard is lacking some details as we didn’t provide a reference model for the courtyard. The more references you add, the more details you get. If we go to the model view, you can even see all the wonderful details for the window recesses. Another exciting feature of this AI model is the ability to create very convincing 3D models from 2D plans. Here I’m using a real plan of an elementary school designed by Daz office. If you include just the plans and no details, make sure you add a clear description of the materials and sizes. Otherwise, you can just add other images to illustrate what materials you want. this case, I’ll add an external image and ask the AI model to transform this into a 3D isometric cutaway plan with a white background using these two as a reference. Surprise me is that it picked up all these voids in the wall and has done a convincing job at adding the furniture. Now you’ll have to take this model with a pinch of salt as you can see that sometimes sections are missing and there are inaccuracies. For example, you can see that some extra rooms have been added where in the plan there should be empty spaces. So you need to pay attention iterate over with the model or use in painting upscaling to fix this. So it is a bit hit and miss sometimes. You can also see that there’s some strange things happening like trees in the wrong places and not all the rooms cut. But this can be easily iterated through. You can just type in the changes you want and the trees are gone and the room is cut through. There are still some number annotations which I don’t want. So I can ask to remove these two. This method also works well for 3D roof pans where you don’t need to cut away and you would like to include some complex forms. Here I’m using the botanical garden design by DARP and I fed the model a similar prompt. The green houses are looking very slick and what’s interesting is that all these have been populated with trees and make it more realistic. The surrounding landscape looks very convincing too. This would be great for architectural diagrams. There’s no need to spend hours photoshopping. Show a few more examples. You can take famous landmarks from around the world and create new models, diagrams, and perspectives with ease. Here, I’ll fly to New York to visit Little Island, spin around, and take a few screenshots. The more references you can give the model, the more accurate the image generations and consistency will be. And since we are New York, I will also go over to the vessel landmark and take some screenshots of this as well. I’ll drop these two screenshots from different angles and ask to create an architectural wooden 3D model from an isometric view. Here we have quite a playful wooden model of little island with these modelike trees included. Banana banana is great at creating different styles while maintaining object consistency. Once again, you can see the accuracy is not at 100%. As you can see, there are some like terrace areas made up in the corner. Then finally, let’s take the free images of the vessel and try to make a realistic cutout model with no background and mess from the Google Earth. And this has actually done an excellent job at removing the background. Even between the staircases, all this mess has been removed. So, it really depends on what input images you use. And finally to show you that this is not restricted to just real plans and image. You can generate your own image or sketch and ask to transform it into a 3D isometric view from above within a cubed volume. You can see it’s taken my prompt and it has transformed that render into a rationalized cubed shape while keeping the main style has all been rationalized. And this is quite easy now to bring over into Hanuman and transform it into 3D. All the organic window recesses have been brought out. This could be a great starting point for the form finding process. From this clean, rationalized view, you can create many more useful drawings. An obvious starting point be to create all the elevations. And look at that. It has cleanly labeled and arranged all the elevations. Only issue is the strange glass windows, but that is because the original reference image actually had these wonky windows and the model was just following my instructions. But these could be removed and commented out. You could also try adding more details such as annotations and levels. However, I found this to be very unsuccessful. No matter how much I prompted it, the numbers didn’t make any sense. Let’s try and create a typical floor plan instead. This is much more successful. It is realistically placed the core and sensibly arranged the rooms, although it is one desk per room. And I could have asked maybe for an open plan layout instead. Although, of course, there are plenty of mistakes within this plan. But you can see that the elevations are showing the wavy facade, but it does keep track of the rigid design. I have just scratched the surface of what is possible with Nano Banana. The next video I will take a look at how you can make full architectural drawings and presentation from just one reference.
22 Comments
I had the same idea to test nano-banana in this way, glad you beat me to it 🙂
Nice!
AI is Very Very dangerous. We just don't know it yet..
Impressive and thanks for your work putting this video together. The technology is pretty mind blowing . . . Good that this has to be done on-line. My 3 hard-drives are absolutely stuffed full of checkpoints, models, Loras, workflows, and python installations. PS: (Comment on modern architectural styles: The more I see the ZahaHadid/Bio-mimetic/pseudo-organic style (impressive though they are to the eye) the more unsustainable, more unaffordable, more expensive to maintain and leak prone such buildings will actually be. 🙂 The future increasingly looks like what it already is – a two tier feudalism between corporate overlords and peasants.
What is highest resolution i can have nano-banana output? i am getting some unusably small files 🙁
I downloaded a Photoshop add-on for this model, a Photoshop script. I hope you talk about it once, it's crazy.🎉🎉🎉🎉🎉🎉
seguiremos en proceso de aprendizaje. se ve una herramienta poderosa
Thanks for sharing 😊🎉❤
Well, many thanks for that explaination, that was insightful. I share to my interns, that opens a wide range of possibilities. Cheers.
Well, many thanks for that explaination, that was insightful. I share to my interns, that opens a wide range of possibilities. Cheers.
오디오 트랙에 한국어를 추가해 주세요~
Well, when you take screenshots in google maps … I mean it's obvious that there is "some" 3d model available already "anywhere", isn't it ?
Thanks for also providing one or two drawbacks, that's solid !
Not free. 1 use test then pay.
I tried the exact same prompt but with a different building (company building of one of my friends) and it didn't work. i didn't get a 3D model but the same stupid screenshot I uploaded….
pure craziness from someone who just finished learning 3d a little while back.
If only I had these tools back in ARch. school… fighting with PS or heavy files exports….I could have saved so much time.
🤯😱
i feel like generating the isomorphic image was not necessary for hunyuan 3d
My God
It looks funny when he takes a screenshot but paste the different picture in the chat 😂
Sorry but this does not work? Used the same prompt and I get the same image output as the uploaded image?? no model like output as per your image generated
Didn't work for me