Playing arround with Sketchup diffusion

jo-ke

and another one sketchup to controlnet- stable diffusion

jo-ke

It's very cool to see the interplay of accurate rendering and stable diffusion. Are you already using this technique for client renders, or is it just experimentation?

Cheers,
Thomas

not yet, but I am trying to get better results with combining different methods. But I am still not happy with it.

Rich O Brien

@jo-ke said:

not yet, but I am trying to get better results with combining different methods. But I am still not happy with it.

Have you tried ComfyUI yet jo-ke? It might help you dial it in but the learning curve is high.

If you are not familiar there is a ton of stuff on Youtube plus theres a great site https://comfyworkflows.com/ where the images shared contain the metadata of how they were made.

jo-ke

yes, but after an update ComyUI does not work anymore. I am more familiar with automatic1111

In the End: I get the best results with renderings. Here: Twinmotion

jo-ke

twinmotion + stablediffusion gives the perfect imperfections...

I love this combination

Rich O Brien

@jo-ke said:

In the End: I get the best results with renderings. Here: Twinmotion

Rolling up your sleeves and getting your hands dirty still looks best. Diffusion can help you in terms of direction to take with lighting and colour palette. But for a final result I still believe you need human creativity.

But AI will get there and it will be interesting to see how the likes of Vray, D5, TwinMotion etc respond.

These shots are though. Really nice output.

L i am

@jo-ke said:

I am testing Sketchup diffusion.
I took an old model and let it go through the new app sketchup diffusion. the result is quite good but the quality looks a bit like a sketch. I took the Picture to stable diffusion and developped it further with image2image and realistic vision. The result is really great...

I find it interesting that the sketchup dufusion and stable diffusion gave a very similar treatment to both images. For instance the sky is almost identical as is the lighting. The only difference I can see is the material mapping and roughness. I wonder why this is?

rv1974

@rich o brien said:

@jo-ke said:

not yet, but I am trying to get better results with combining different methods. But I am still not happy with it.

Have you tried ComfyUI yet jo-ke? It might help you dial it in but the learning curve is high.

If you are not familiar there is a ton of stuff on Youtube plus theres a great site https://comfyworkflows.com/ where the images shared contain the metadata of how they were made.

What are benefits of Comfy (in comparison to A1111)?

rv1974

@solo said:

So, this whole AI thing may be more than a fad?

So far the only use (from all this AI craze) I see is 3D people enhancement in stills.

rv1974

@rich o brien said:

@jo-ke said:

In the End: I get the best results with renderings. Here: Twinmotion

Rolling up your sleeves and getting your hands dirty still looks best. Diffusion can help you in terms of direction to take with lighting and colour palette. But for a final result I still believe you need human creativity.

But AI will get there and it will be interesting to see how the likes of Vray, D5, TwinMotion etc respond.

These shots are though. Really nice output.

Meanwhile Chaos implemented DLSS 3.5 thing in Vantage - and it it really makes wonders

Rich O Brien

@rv1974 said:

What are benefits of Comfy (in comparison to A1111)?

Its node based so its easier to follow logic if your comfortable in node based environments.

If not stick with A1111.

jo-ke

@l i am said:

I find it interesting that the sketchup dufusion and stable diffusion gave a very similar treatment to both images. For instance the sky is almost identical as is the lighting. The only difference I can see is the material mapping and roughness. I wonder why this is?

The Sketchup Diffusion Output is the Stable diffusion input. That is the reason both look similar but stable diffusion img2img devellops it a bit further to a better result.

rv1974

interesting observation on AI:

jo-ke

overlay in photoshop:
-sketchtup
-skectchupdiffusion
-rayscaper

nice result

pixelcruncher

@jo-ke said:

overlay in photoshop:
-sketchtup
-skectchupdiffusion
-rayscaper

nice result

Really nice! (I’m sorry, I accidentally downvoted, the buttons are too small for my thick thumbs).

jo-ke

further development in stable diffusion

deutsche-reihenhaus-165 20240316 22h58m52s.png

sketchupdiffusion

jo-ke

I am using sketchup diffusion to build textures of buildings.

At first I am looking for sketches of fassades.

then I import it to Sketchup, define the colors in the promt and render it.

then I project the texture on my object and model the rest.

This is very fast and effective.

Untitled 20240401 00h08m25s.png

jo-ke

a funny weekend project.

some years ago, I was 3D scanned and they made a small 3D Modell of my body.

I scaned the small figure in3D and made a Sketchup modell of it.

The I renderd it in Sketchup diffusion and put it in stable diffusion img2img.
upscaling with my Lora and a reActor Faceswap.

nice result....

su view:

su diffusion:

Untitled 20240420 18h23m32s.png

stable diffusion:

jo-ke

quick tutorial to show how I use Sketchup Diffusion:

At first look for a drawing of a building, then texture a cube with it, and set the view to front view. then start Sketchup Diffusion. I typed for promt: "fassade of an old building, dirty beige sandstone" and hit "generate".

then I took the best one and hit "add scene". then I took the sketchup function "project texture" to exchange the drawing with the AI texture.

the cube is textured

some work on geometrie.... finished in less then 5 minutes....

and a quick render in Twilight Render

jo-ke

playing some more.

again the 3D figures...

su view

sketchup diffusion to stablediffusion, lora and ReActor Faceswap

not to bad