Hello everyone!
I'm excited to share with you my new 7Lambda_Negative_v.1!

In this LoRA, I've carefully selected and compiled the most problematic issues typically encountered during image generation. After meticulous dataset curation and training, I noticed a significant enhancement in image quality, even without using positive LoRAs!

How to Use

It's super simple, just two easy steps:

Step 1: Add the Load LoRA node in ComfyUI with these recommended settings:

LoRA loader settings:

• strength_model: -1

• strength_clip: 1

Step 2: Include (super-wrong:1) in your negative prompt. Here's an example of my negative prompt for guidance:

(super-wrong:1), ugly breast, ugly soles, ugly foot, bad nails, deformed nails, bad anatomy, bad hands, ugly hand, (plastic_skinn:0.7), bad boobs, bad breast, (3D:0.6), horrible, terrible, simplified, abstract, artificial, ugly, duplicate, morphed, chaotic, text, watermark, typo, bad_quality, worst quality, worst detail, low quality, normal_quality, poor_quality, low_resolution, (bad legs anatomy:1)

Feel free to experiment with values slightly different from my recommendations for finer adjustments:

• strength_model: from -1 to -0.1

• (super-wrong): from 0.1 to 1

Compatibility

This LoRA is specifically trained for use with SDXL models and was trained on the illustrious model. It significantly enhances image quality for both SFW and NSFW content, Anime, Photo or another you own style!

Enjoy creating, have fun experimenting, and please support my research if you find this useful!

Cheers, 7Lambda!