Experimenting With Automated Procedural Texture Generation — I’d Love Your Feedback!

General Forums Off-topic Chat
1

Hey There!

I’ve been tinkering with a little side‑project lately, and I think some of you might find it interesting — especially if you enjoy procedural workflows, texture creation, or just exploring weird technical experiments.

A couple of weeks ago I had one of those “wait… what if?” moments.

The idea was simple: Could we approximate a raster texture automatically using a combination of procedural textures? What if there was a system that takes your 2K texture and says “okay, you need a noise texture, make it through a clamp, then add a voronoi texture to it” to get really close to your image texture?

So… my proto/demo/proof of concept does this (or at least tries to do so)

I started experimenting, and after many tests (and many crashes), I now have a very early prototype that actually works well enough to share.

It’s not magic, it’s not production‑ready, and it’s definitely not fast — but it’s a start.

If you’re curious, you can check out the prototype on GitHub: GligDav/ProcTexLab. Clone the repo, follow the install instructions, and you’re good to go.

This is very much a “baby’s first steps” situation, so here are some honest notes on what you should expect (and not expect):

  • Performance is rough. Even with GPU acceleration (CuPy), fitting a texture can take minutes or hours. If you plan to import results into Blender, keep the settings low.

  • The Blender addon is fragile. It breaks easily. I’m actively working on stabilizing it.

  • “Blurry” results are expected. This is an approximation tool — many raster textures simply can’t be recreated 1:1 with procedural building blocks.

  • Crashes and bugs will happen. It’s early research, not a polished tool.

Since this community is full of artists and technical creators, I’d love to hear what you think:

  • Would you use a production‑ready version of this?

  • Do you see potential in the idea?

  • What would you change or improve?

  • Are there use cases I might be overlooking?

Even a short comment would help me understand whether this project is worth pushing further.

:folded_hands: Thanks for taking the time to check this out!

I’m excited to hear your thoughts — whether it’s “this is cool,” “this is weird,” or “here’s how you could make it better.”

— Dávid

PS: unfortunately, I cannot add attachements to this post as my profile is new, but here is a quick YouTube video showing how the thing works: https://youtu.be/K4TGJh329aQ