I have a white ceiling and it makes it wood, etc
That shouldn’t be too surprising. Diffusion does not exactly follow the materials in your model. Did you tell Diffusion you want a white ceiling?
This was my first time using diffusion. While I understand AI works from text based prompts, I would assume at minimum, it would take the materials/colors already in the base sketchup image and go from there. I will try one where I describe all the materials and see how that goes.
Unfortunately your assumption isn’t correct. The AI makes many assumptions. You can steer it by giving more specific prompts but it will still make its own decisions. Note that Diffusion is not supposed to be an alternative to rendering applications like Vray.
It looks like diffusion isn’t the right tool for you, at least for now, if you need a rendering engine that creates images faithful to your model, there are a lot of engines that work either inside of sketchup or have a direct link with sketchup. Vray and enscape are the ones I’m currently using, enscape for videos and vray for still more realistic images.
Yes. Currently use enscape
I use Diffusion mainly as an inspiration source, probably in the future it could be able to generate more accurate images. There are engines that use AI that can generate more faithful to the model renders, Veras is a pretty good one, it’s paid though.
Diffusion is an advanced toy.
1 Like
It would be improved if “using existing materials” instruction worked.
1 Like
I think we understand that Diffusion works differently than rendering but this being an extension of SU, it would be extremely helpful, if not necessary, for the AI to take materials into account. We are very early into this, and it is interesting so far, but the discussion should continue to be, “how can diffusion be improved to better assist the user”.
If I understand you correctly, you want diffusion to be a rendering extension.
Not want, but that is how it is being marketed on their website which shows a basic sketchup model and then highly rendered views of the same model via diffusion.