Blender+ SDXL + comfyUI = fully open source texturing workflow: 1. cameras in blender 2. render depth, edge and albedo map 3. In comfyUI use control nets to generate texture from view 5. project back and blend based on confidence (surface normal is a good indicator)

Example, here are the three maps rendered from blender followed by the frontal view that SDXL renders (the comfy workflow is in the meta data of the image in case you use it)

when projected back, you get this. This is just one view, so to fully get the model you would need more views while keeping the seed the same. You can combine it with inpainting to only effect the not texture areas and give context to what is made so far.

generating each view takes only a couple of second on a good gpu, so you can be done with modeling in a couple of mintues

Especially in Blender, AI solves all the problems we don't have. Let's have AI do the inbetweens, or do samples "upscaling", maybe remeshing, or create the basic geometry of a complex shape that an artist can use as a base. Creating textures is the least of the issues

You might not have texture fixing problems, but I am glad that when fixing the texture of a random surface of a photogrammetry mesh, I can be done with the task in 10 minutes instead of 50 😅 The other problems are also important, you should give solving them a try!

I came from our reddit conversation superstarbootlegs, to here 😀

I wonder if, in some cases, running a segmentation pass using tools like SAMPart3D or HoloPart could be beneficial. These tools could help by creating polygroups or material slots for different regions of a model.