I have trained 300+ trainings of LORA, DORA and LyCORIS and this LORA is by far my best.

98% Success rate up-scaling feminine models with a bias towards ???? females.

This model tends to take off clothing as it was trained on 100+ images of females flashing breast, vulva and public exhibitionism. By chance I found that on nearly every image it had high detail at up-scaling 512x512 to 1024x1024 and then to 1536x1536

The Zipfile has an example workflow as while I'm sure the jpeg save Civitai a ton of space they do not import. (Also Civitai the Downsized images look like garbage)

Steps are first generate your 512x512, scale by 2x, then scale the 1024x1024 by 1.5x

from their you can even go to 1792x1792 at 0.2 (See table for CFG and LORA values)

• For 512x512 generation:

• CFG 5-8 with Karras/Euler recommended. (Lora Value 0.5-1.0)

• For up-scaling the 512x512 Generations:

• CFG 4-5 Karras/dpmpp_3m_sde_gpu (Lora Value 1.0 CLIP SET AT 2.0)

• De-noise at 0.5 to as high as 0.7 (No High Rezfix needed)

• For up-scaling the 1024x1024 to 1536x1536:

• CFG 4-5 Karras/dpmpp_3m_sde_gpu (Lora Value 1.0 CLIP SET AT 2.0)

• De-noise at 0.4 to as high as 0.5 (No High Rezfix needed)

This model also excels at public ??????, recommend to add ???? to the negatives list and use terms such as:

• In Public, exposed, open legs, ???? peek, ?????? peek.

IMPORTANT: Leave your long list of negative behind. Minimal negative prompting is needed.