BeautyFoolXL2025

v2: The v2 model is the most consistent, flexible, and realistic model I have achieved so far. In many cases, it produces better results than FLUX models. However, there may still be issues with fingers and nails. For this, you can use LORA plugins that focus on better hands. If you add one or more of the negative embeddings listed below to the negative request section, you can achieve some minor improvements.

step: 30-40

cfg: 4-5

sampler: dpm++ 2M/3M SDE

scheduler: Karras / SGM Uniform

VAE is included, but if you think the colors are too dull, you can use any sdxl VAE of your choice.

/model/139688/imgfixer-or-negative-ti?modelVersionId=159184

/model/1070219/sdxl-negative?modelVersionId=1201288

/model/1531979/cyberrealistic-negative-sdxl

If you are using SwarmUI, add the following to the end of the positive prompt to obtain better-looking faces: segment:face,0.5,0.4

If you are using Automatic1111, I recommend using ADetailer.

v1: BeautyFoolXL2025 is a model that succeeds in photorealistic images and has a NSFW tendency. It was created by combining more than 10 models in certain proportions.

Vae is not included. Please use in combination with a Vae.

Usage Tips:

I recommend typing the usual prompts such as "realistic, analog photo, 4k, 2160p, score_9", among the positive prompts.

Step: 30-50

CFG: 4-5

Sampler: DPM++ 2M/3M SDE Karras

In SwarmUI with RTX3090 Graphics card, it takes 31-32 seconds to render an 832x1216 resolution image (step:40) with 2x upscaling (Refiner Control Percentage:0.2 and step:25) and face enhancement.

If you use this model with Flux as a Refiner (Upscaling) model, you can get high quality NSFW images that Flux alone cannot handle.

Or you can convert previously created 2D, 3D images into photorealistic images using the img2img method (Init Image Creativity: 0.45-0.65).

Note: No lora, embedding, addon or inpainting is used in shared images