Created using AbyssOrangeMix2_hard, the model consists of 357 high-quality images and has undergone 9,750 training steps. It's my first project, so it may not be perfect for all styles or situations, but overall should work well with the right keywords.
For medium-quality images, utilizing a single checkpoint will be enough. However, if you're seeking optimal results, follow these two steps:
Process the model through AOM or AOM2Hard with LoRA strength set between 0.6 and 0.8.
Input the output into another KSampler (preferably AOM2Hard) and set the second LoRA strength to 0.4. (The example images provided were created using this method.)
Recommended VAEs are AOM2Hard and vae-ft-mse-840000-ema-pruned.
Feel free to reach out with any questions or suggestions you may have.
If you want to boost my workflow for these cultured models with a quick cup of coffee, you can jump over to Ko-Fi: https://ko-fi.com/culturedaiuser