Hi guys.

As stated in the first image, this lora is built on the Kroma v0.2 turbo finetune so download that first to get it to work properly. I opted for Kroma because it partially de-distills Krea, increasing its scope and allowing the model to better learn different characters. The model can be found in the link below. Please support Lodestone and the work he's doing.

https://huggingface.co/lodestones/Kroma/tree/main

For usage, my workflows are attached to each showcase image, but here is the general guide

• 1.2 cfg, 15-20 steps, res_2s/beta (not fast but good results)

• Refer to the subject as Tokine and use natural language prompts

• I begin each prompt with "An anime scene of" and end with "The image is a still frame taken from a high quality animation" to get the anime screenshot feel.

The intent behind this lora is to produce results as close to believable screenshots from the actual show as possible. Let me know if you have any questions or feedback.