Native ArchiCAD AI Visualization Plugin - Seeking Feedback

[mike_linespire]

Hey everyone,

I've built an AI archviz addon for ArchiCAD that is using the latest Google models. As primarily a building designer, I wanted an addon that streamlined creating the image directly in ArchiCAD and placing it on the coversheet layout for our projects without needing to export/import.

I've played around with these AI models from the start and always been disappointed that whilst they were impressive, they weren't accurate enough to share with clients (missing windows, roofline changes, wrong materials). The latest models from Google are the first time I've felt they are producing client ready images (9 times out of 10!).

I'd love some feedback on the tool so I've setup a free trial that includes 5 image generations (reach out and I can top you up with a couple of extras if you need them). I'm going to get a macOS version going as soon as I can too.

There is a before/after slider on the website where you can see how accurate the AI model is at matching the geometry of the ArchiCAD model

Operating system used: Windows

[Botonis]

Always intersted.   I really would love to give it a try!

Are you referring to Gemini?

[mike_linespire]

Hey Botonis,

Yep its using Gemini 3 Pro Image (nano banana pro) at the moment which I've been really impressed with. I've set up the plugin so I can always swap in whatever AI model is working best for archiviz at the time.

Let me know if you give it a go!

https://www.linespirereflect.com/

[PB]

Hi Mike,

Most promising. Please let us know when you have a Mac version to try out.

Some questions: (from an AI neophyte)

1) Can you "lock-in" the elements added by AI so that they are consistent through different views?

2) Can the process insert a design into an existing site? (from photos, aerial images, surveyors data, etc.?)

3) Could the process be adapted to upload a 3D model and generate views & video rather than working directly from a fixed 3D view?

4) your thoughts on pricing for those of us who may not need it on a regular basis?

Many thanks

P.S. How is copyright, privacy, and intellectual ownership handled? (from the Google side - your terms are clear on your site)

[rkres]

I'll also give it a shot, thank you for sharing your work.

[rkres]

I second the option to lock in elements, for more consistend shots.

[mike_linespire]

Hey PB,

These are great questions!

1) This is a tricky one and something I've had to battle with in my own projects. The AI can be very consistent if it recognises the Archicad element and its association in the prompt is strict. So for example if my Archicad model had a vertical weatherboard surface and my prompt said "James Hardie Axon 133mm in Dulux Vivid White", it should be able to keep that material consistent in 2 different images. Where you would definitely have problems is with elements added entirely by AI. For example if you were generating an image of a pool area from 2 different angles and prompted "add floating pool toy", you would likely end up with totally different pool toys in each image. If you need something to remain consistent across both images its best to add a 3D element to the model that the AI model can reference and then be specific about it in the prompt. 

2) AI can definitely do this, Gemini 3 Pro Image can do this particularly well. It's not something you could do from my addon yet though because you'd need to be able to add reference images which isn't possible yet. Its a great idea for a future feature I can work on.

3) I'm not sure I understand exactly what you mean but a feature I'm looking into being able to work with archicad camera views instead of just capturing the 3d window. That way you could setup all of your camera views in advance and then process them together with the same prompt.

4) I'm definitely going to think more about pricing in the future. I think it would make sense to have a cheaper tier with less credits or a longer term tier for people that need it more infrequently.

Google's privacy policy is actually quite positive. They claim they will not use the data to train any AI/ML models or retain the data (subject to some specific terms) . I made sure it was clear on my website which API we're using because I think its really important that people know how their intellectual property is being processed. Here is a link directly to their privacy policy if you're interested: https://docs.cloud.google.com/vertex-ai/generative-ai/docs/vertex-ai-zero-data-retention

Thanks again for the questions. The macOS version is top of my list and hopefully won't be too far away. I'll keep you posted

[mike_linespire]

Hey rkres,

I mentioned this above to PB but something you can try is adding a 3D element in the Archicad model of what you want to stay consistent and specifically mention it in your prompt for both images. 

For example if your prompt includes a feature tree in the front yard, add a tree model where you want it to be and then in your prompt say something like "tree in front yard is a mature olive tree"

Consistency is definitely the hardest challenge with AI generations though. Let me know how you go!

Thanks

[rkres]

Thank you for the advice, that makes sense. 

[PB]

you for your clarifications, Mike. I look forward to the Mac version.

On my third question, I'm trying to ascertain two things - if you can help:

a) Your example shows the landscaping (driveway, planting, etc.) into which your model is integrated. As & when the AI comes up with a good design proposal, ideally we could then instruct that the design elements visible - including the 100% AI items - are kept for following renders or iterations. So , for example, the driveway is an attractive design, hence we would like it to be exactly the same on the next render which is from the other corner of the house from which the drive is still visible.

b) By using a 3D model (native PLA, IFC, Datasmith, 3DS, etc.) is it currently possible for the AI to "adopt" the model and create its set of textures, setting, people, vegetation, etc. so that a consistent set of stills & video can be generated from it. I'm guessing not by your response, and from what I have seen on the internet, although I suspect we can't be too far away from that. My reasoning being that if the 3D model is treated by AI in the same way that an Enscape ot Twinmotion model is created, we then have consistency across renders & video. Perhaps this is something which would need to be developed by Chaos or Unreal to anchor the AI input to the 3D model?

Thanks again for sharing your work and I wish you every success.