!!! LoRA can be difficult to use !!!

This is my first attempt at creating LoRAs, so they aren't perfect. They have some flaws and might be a bit unstable, but you can still use them.

On this page, I will share various styles that I use in my own works.

You can find more details about each LoRA in the “About this version” section.

WAI/ILLUSTRIOUS version is here.

Many thanks to @wwHEY — I couldn’t have done it without her advice!

But she's not responsible for the crappy quality of my LoRAs. x)

描述:

This model provides decent aesthetics and depth, however..... it is VERY difficult to use. I don't know what's wrong with my models, they always turn out like this, but... yeah. :(

Trained on artworks of artist kan liu (666k). I edited all dataset in photoshop to improve the quality and effects.

A few tips on how I USE IT. You might do better in a different workflow. Feel free to experiment.

• Using artist tags will stabilize the generation greatly. It is better to write them at the end of your prompt. For some reason they work better that way. The kan liu \(666k\) tag is for more similarity to the artist's style. Also you can use stabilizer LoRAs, there are plenty of them already.

• 2) Use a klskx tag with a weight of 1.1 for better result (in my workflow it is better smh)

• 3) The recommended weight of the LoRA is something I haven't fully determined myself yet. Currently using 0.75-0.85. Experiment.

• 4) "blurry background, depth of field, bokeh" - to create depth, this LoRA does a great job with them.

• 5) "dynamic pose, dutch angle" - more dynamics.

• 6) "outdoors, day, sunrays, sunlight" - for better lighting. I will try more tags later to update this section.

• 7) Sometimes ‘hydrokinesis’ and ‘water’ appear. Just add them to the negative prompt. Or in the positive if you want to see it in the picture.

I'll probably retrain this model again, but for now, I will keep it like this.

训练词语: (klskx:1.1)