designing with self trained minimalistic LORA models

minimalistic dataset made from handdrawn abstract drawings

With KOHYA, we have an open source tool, to train our own LORA models on our home machines. This can be done with even a very small amount of image data. This experiment is a proof of concept to show the limitations of such minimal AI models. You can try it on your own if you download the pretty small LORA: trx_one

(hyper detailed breaking ocean waves, gusty sea :1.2) , trx_one thin curved black ink strokes, (white plain paper background:1.2), high contrast, sharp, diffuse bright lighting, (blur, blurry:1.1), dots


training with two colored pen drawing patterns

hand drawn minimal training set with two colored pen drawings
drawing a complex motif – pen drawing – with strong LORA influence
generating a complex embroidery with mid LORA impact

training with hand drawn color panels


handdrawn color panels as training dataset
using the painterly LORA to guide the img2img overworking
using the painterly LORA to guide different forms like textile cables
using the LORA to guide form finding process – here to set cables in a composition
using the LORA to guide form finding process – here to set textile cables in a composition

training with plain photography

minimal wild wood training set in 512×512
These slide show the variation of LORA strength appied to a fixed seed
using this wood structured LORA to modify water is quiet ineresting
with more strength, we can create more painterly effects

sources and help

Great simple guide how to install Kohya on your machine!

Good guide how to use Kohya to train your model!

(SPOILER: Check the structure of your training folder! )