INSTRUCTIONS AND README,
Please for those using ComfyUI the main model file is “flux_PicPong-Prime.safetensors” goes into the Comfy 'models' folder, inside the 'unet' folder.
The other two, the “flux_PicPong-Prime_clip_lsafetensors” and “flux_PicPong-Prime_t5xxl_lsafetensors” they go in the 'clip' folder;
as for the ae.sft which is the VAE, it goes in the 'vae' folder
The model was trained on FLUX.1-dev vanilla using custom made training scripts so that CLIP_L and T5XXL may be trained in order to introduce new concepts and overcome the censorship, therefore the model will not work without them.
The trigger word is very important and should not miss from your prompt.
Here's a list of all characters trained in the model:
Character_sheet
The flux so called "censorship" has also been removed.
The model can create NSFW content and currently contains close to 100 distinct characters.
The style of the model was inspired by PicPong, which I invite you to read more in the about section on CIVITAI.
Given this is a FLUX DEV model, it still abides by their licence meaning you may not use it for commercial use.
I hope you will have fun with this model and if you feel like helping out our friend that made the whole dataset possible, and alone; I just tagged and trained the model, and you want us to make improvements; compute is the key that will get us there.
We can train LEWD with COMPUTE All the images in the
Metadata_Samples folder have been generated with ComfyUI and contain everything required to run the model.
All you have to do is input your own prompt, or change an existing one.
