LoRA trained on 386 Images of Diffuse Texture from polyhaven and manually download and describe tags for each one of them... while respect polyhaven tags.
I shared my dreambooth config and the Training Data so that you can reproduce this LoRA, sharing also my process to train my model BerryGold on Large amount of anime images.
Note
SD1.5 can already create texture but not without shadow/reflection.
Tips
Related models
Checkpoint base on SD2.1 Texture Hell, Texture Diffusion. (withpoly, Text2Tex 1.), ESRGAN
Description
Tags list for v1.0 Experimental_test_2K
This checkpoint is meant to generate 2048x2048 images without upscaling! NOT 768x768
not a LoRA, but is based on checkpoint 2.1-768, and trained on images with a resolution of 2048x2048.
This is a proof of concept for comparing high-quality datasets to 512x512; in the end, 512x512 remains superior.
Added 16 textures.
Added Trigger Words: ", seamless, shadowless" Will be removed in next version.
Fixed tag list issues by normalizing less common tag variations, along with additional miscellaneous dataset fixes.
FAQ
Comments (3)
8 GB Lora ?
@DukeDarkShadow
About this version:
This checkpoint is meant to generate 2048x2048 images without upscaling! NOT 768x768
not a LoRA...
ignore v1.0 Experimental_test_2K as it is a TEST/proof of concept, just use v1.0.
This really screws up caching services that think this file is a LORA.