Replicates the style of AI-artist <a rel="ugc" href="https://x.com/748cm">748cm</a>. The artist has many different style, so we might not have the same expectation of the LoRA. This is mostly trained on the more line heavy illustration (though there is some painterly ones in the dataset). Could be interesting to have one trained purely on the painterly ones, or purely on the line heavy to have a stronger specialized effect.You can push lora strength up to <code>1.3</code> if you want stronger effects, after that anatomy starts to worsen from my testingAdd keywords like <code>gritty</code> <code>noisy</code> <code>lineart</code> <code>painterly</code> <code>washed colors</code> <code>glitch</code> etc, to steer the model towards the style you prefer.

Style > 748cm

Training Settings: batch size 5, 1024px, ~190 images per epoch, optimizer is <a target="_blank" rel="ugc" href="https://github.com/Koratahiu/Advanced_Optimizers/tree/goemtric_wd">SinkSGD_adv</a>

Files with SHA256: c237c2c1598d92a185ea1cb5da60f51f7ff556291b220e414506473ef6b013d6