A special blend of herbs and spices to tantalize your taste buds. These models are flux dev based models and are merged with a number of loras in an attempt to maximize nsfw photorealistic images of beautiful nude women with beautiful breasts, and with the addition of version 1.4, beautiful vaginas. While they are primarily focused on upper body portraits, they are capable of generating full body anatomically correct female nudes. Lora merge ratios were chosen to minimize bleed through when prompted with clothing. Each version is an individual mix of loras, they are not improvements of each other. Each model has its own strengths and weaknesses. A special thanks to the lora owners who allowed me to include their work in these merges. I hope you enjoy them.
NOTE: When ByteDance Hyper-FLUX Acceleration LoRA is merged into a model and the model is converted into fp8 precision, the conversion process screws-up the hyper 8step acceleration blocks from the lora. This causes the model to behave like a standard (non-hyper) fp8 model. The good news is it can still be included in the image prompt and still get the benefit of hyper 8step with fp8 model.
These are merged models of a base model flux1-dev fp16 and combinations and ratios of the following loras :
https://civarchive.com/models/667086?modelVersionId=746602
https://civarchive.com/models/681562?modelVersionId=762861
https://civarchive.com/models/652791/jibs-flux-nipple-fix
https://civarchive.com/models/660029/nakednudeforfluxsevenof9nsfw
https://civarchive.com/models/704013/photorealistic-nsfw
https://civarchive.com/models/983498/pussydiffusion-flux
To convert and merge, I used scripts from:
https://github.com/kohya-ss/sd-scripts/
To convert to gguf images, I used scripts from:
https://github.com/Zuntan03/EasyForge/
no keywords.
Description
same content as v1.1 saved in fp8 precision.
FAQ
Comments (2)
Am I the only one who feels this way? FB8 is more realistic
i personally prefer fp16 models. i was running it on an rtx 3060 with 12GB of vram. as far as i know, the gguf models are a statement of compression, not precision. so the gguf model were built based on fp16 model. so, theoretically they should more closely match output of fp16. the fp8 is an actual conversion of 16bit floating point values to 8 bit floating point values. i think it affects the image 'content/accuracy' not necessarily the image quality. image quality is defined by generation parameters. just my 2 cents... ymmv.