This Lora will generate a fully artificially generated consistent character of a next-door type, tall and skinny redhead woman in her mid-twenties. She might not be the prettiest girl in the universe (intentionally so) - but she has green eyes and freckles, despite being created following my Simple SDXL Consistent Character Generation using just Forge and the Civitai Trainer workflow (she has a cute smile, though).

To learn how I solved the issues with eye color and freckles in that workflow, please read the paragraph about dealing with freckles and eye colors in my article.

She is also fully NSFW capable (provided your base model can do NSFW), however, you'll have to prompt for nude body details such as pubic hair, as she has only been trained on clothed pictures (the reason is explained here).

Training data and example pictures for SDXL have been generated using the excellent LUSTIFY! NSFW checkpoint, example pictures for FLUX have been generated with the FLUX.1 D base checkpoint.

Usage and Prompting

FLUX:
Drop the lora in (the default strength of 1.0 is fine for the FLUX base checkpoint) and ask for a picture of BethanyAI. Please be aware that this LoRA works best with the original FLUX.1 D checkpoint, I have not received very good results with finetuned checkpoints even with a strength of 2.0 (somehow, FLUX finetunes seem to struggle with LoRAs trained on the base checkpoint). Contrary to the SDXL version, Bethany is fully NSFW-enabled on FLUX, however, as FLUX base really isn't good at nudes, don't expect exceptional results.

SDXL:
Drop the Lora in (at a strength of about 0.8) and prompt for BethanyAI, clothes, poses etc. If you'd like more freckles, feel free to additionally prompt for freckles or include body freckles to your liking.

As stated before, when generating nudes, be sure to also prompt for pubic hair and similar nude body features if desired.

Be aware that her freckles may sometimes bleed through her clothing, in this case I advise to prompt for clothes before prompting for freckles (no guarantee, though).

As this Lora is primarily a proof of concept, I did not particularly pay attention to versatility, particularly when it comes to different backgrounds. If you keep getting a white wall as a background, negative prompt "simple background", as most training images had a uniform background, but this might not help in all situations. At some point, I might actually create a second version that will be more versatile than this one.