How is Archicad's integrated AI?
We participated in a public webinar organized by Graphisoft. It was very interesting and we learned a lot:
- At its core, this AI has a Stable Diffusion algorithm and it was trained to generate building-specific images
- The process is that the AI takes a snapshot of the 3D model and that’s what it works on, instead of the whole model, that’s why it is recommended to position the model at an angle in which it can easily recognize the shapes.
- This trained AI essentially understands five input categories: Visual style, Architectural style, Building type, Materials, and Environment.
- The AI was trained on 1024×1024 & 1216×832 image sizes, Graphisoft says that it is possible to get better results if these are used, but during our testing we didn’t notice any differences.
They also mentioned that it can help if you fill out the entire screen with the model. It can recognize materials as well, so it’s not mandatory to make a grey model, feel free to use materials on the surfaces. As we learned, the engine doesn’t learn from the prompts but it’s recommended to restart the plugin now and then.
There are also some hidden features, for example, if you write #devModeOn and hit the ‘Generate’ button it will deactivate the trained AI and now you can generate in theory ‘skiing ducks’ or whatever you want. To deactivate it just insert #devModeOff. Another hidden feature is if you do a shift+[Generate] it will loop back and use your last snapshot for the base.
The Testing Phase
We put to test the new AI Visualizer plugin from Archicad and here are some of our observations:
It’s a great tool for experimentation but don’t expect it to generate mind-blowing renderings from your detailed project. In terms of this, it cannot compete with rendering programs such as Twinmotion or Lumion. During the concept design phase, it can be helpful; however, we would not use it to present the final project.
The results of our visualization test for the Concordia Hotel in Târgu Mureș, Romania:
Comfort vs. Results
This particular plugin creates 3D visualizations within the Archicad software, which I think we can all agree on, is a lot more comfortable since you can access it faster and more easily.
Now the question remains: Is the comfort of the integrated Archicad AI Visualizer more important, than using a different, perhaps more evolved program or plugin, and getting better results?
It’s still in the early stages of image generation from a 3D model and it cannot compete with the rest of the AI programs out there, such as Midjourney or Adobe Firefly. It is a great tool to test and play around with, in certain phases of the project it can be helpful as well. For example when you have to test out an idea real quick, or to test multiple ideas fast. It can also help in finding inspiration.
The results of our visualization test for the Wesselényi Castle in Hodod, Romania:
Would we use it in the future?
At the moment we do not think that it’s worth using instead of a better-performing standalone rendering solution or a render plugin for Archicad, as there are some great solutions on the market such as Twinmotion, Enscape, not to mention professional grade render software/motors such as Vray or Corona. Graphisoft is going to improve this plugin in the future, however, for the time being, maybe it’s not such a great idea to give up on better results for an easier, more comfortable solution.
We do realize that it is still in its early stages, and it has potential, so we will be on the lookout for any improvements.
About the author
