So I ask myself, which workflow / possibilities can we expect in the future from Apps like Enscape/Vray? Or could it even happen that Apps like these become obsolete in the near future an AI will take over?
What are your opinions? What are your expectations?
I have the same opinion as you that Applications like Enscape/Vray will be gradually replaced in the future due to AI. But AI cannot completely replace those apps as they will also be updated and upgraded
So far the AI ‘renders’ I’ve seen online have been ‘like’ the model provided but NOT exactly the same.
If I was to submit anything like that, no matter how good it looked I’d be asked (and probably not politely) “Why did I change the roof design?” or “Why are the windows all wrong?” or “Why isn’t there any gutters and why does the downpipe go diagonally” etc etc
What’s more, If I was to request AI to generate 5/10/howevermany views of the same model I almost guarantee they’d be quite different content/styles. Additionally, coming back to it in 6 months to make ‘slight amendments’ would be a complete nightmare.
The ‘future’ of architectural visualization may be AI but certainly not the present nor do I suspect the near present.
Paul, I think it doesn´t take long, till it´s possible, to upload an own rough model or even a 2d drawing, which can be illustrated by AI. With full control of all surfaces, lighting etc.
I have been expermenting in AI in architecture for the last year, I have tried MidJourney, Versa, Chat GPT, and now Fire Fly. In my opinion the only two that I have found viable is Fire Fly and a little bit of MidJourney. While Fire Fly still has it’s quirks it does somethings really well especially when asking it to do background assets, ones not in focus. MidJourney on the other hand I have used for brain storming and programing, sometimes using it to try and imagine a theme in my head before attempting the render. I haven’t tried control net, but it seems to be a somewhat viable option in the industry.
Where I really think AI will improve our workflow sooner then we think is AI generated 3D objects, I can see this coming fast to the industry and be the norm for populating your scenes. No more searching through countless websites to find the perfect golfcart model when you can just ask for it to be created. I think the first application to go full speed into this industry will have the upper hand by the end, though I don’t see the need for rendering applications disappearinganytime soon. Especially due to some of the unexpected results I have encountered in my studies.
Meanwhile I did some testing with AI Renderings myself. My Goal was, to get realistic Renderings out of the Screenshots, taken from my modeling software. These are the results with StableDiffusion/ControlNet. The workflow with this programs is to mix prompts with drawings.
As you can see, the results are not very accurate. And it is hard to get control over details. But it´s impressive anyway. So now I wonder how (or if) Enscape will implement AI in the near future.
At first glance the interior looks respectable, then I realized the flooring changes direction from one side of the room to the other and it looks like the wood paneled wall was completely ignored. That seems to be the problem (for the foreseeable future) is that AI will guess at something that should be precise. Doesn’t work too great for architecture.
Yes, but I just read some news about that AI will be trained in spatial awareness. I think this is the problem with AI at the moment: AI does not know what´s it´s doing relating to a 3d space.
I think that in the future we will have to understand the expectations of AI rendering programs differently.
1.) We will continue to have rendering programs. AI will certainly be able to help with this even more than now. The programs will have even higher requirements for quality, realism, video and rendering time, etc. and simplify operation. At the same time, it includes the controllability of the render images that we have imagined as part of our ideas all these years when we started rendering the first images.
Therefore, AI alone will never be able to implement our idea of accuracy and ideas in the rendering program. Also, there is a lack of targeted control over reality, detailed images, their environment and the knowledge of feasibility of an architect and interior designer.
We are probably standing in our own way and hoping for the legend of the Holy Grail of AI development. However, the future will be AI-specific and will find another option.
A real way into the future with AI for rendering programs is currently shown by the example: D5 Render 2.9 (see YouTube)
2.) AI rendering programs alone, for example:
SketcUp and Ai, Fabrie AI etc. are uncontrollable tools and will have to be able to do more in the future, therefore unusable
Of course there are AI rendering programs that can create impressive images of fantasy worlds based on a Sketchup cube. (see my quickly drawn Sketchup and the resulting example image below)
Except for the initial glance it’s not very convincing. The minute you start to look at details, you realize just how sloppy AI generated images are. It’s architecture, but slightly melted. To turn an aphorism on its head, AI is not in the details.
