iFAP - SC - RAW

I am putting this into the wild for use by model merge authors to give them another tool in the toolbelt. This is a finetune of my custom dataset of primarily, but not exclusively pornographic photographs of women trained against SD1.5prunedEMA base model for maximum compatibility for merging and lora extraction purposes.

This is not a merged model. It's a raw finetune, warts and all.

On it's own, this model will not produce consistent quality output.

When merged with other quality models, it can provide enhancement in a number of areas, both SFW and NSFW, primarily in the area of skin details and realism for human subjects. You can find further details on the ifap-sc page, as well as caption data for prompt direction and to give model authors an idea of what this thing does.

Here's some training details, as I will likely release other raw finetunes in the future that were trained on other base models and used different training parameters, which produce different results. This should help differentiate them as they become available.

trained on OneTrainer, 2500 images, 8e-7 LR , linear scheduler, 60 epochs, EMA enabled @ 5 step interval updates. Multi-res trained at 512, 576, 640, 704 pixels.

I politely ask that if my work finds it's way into one of your merges, if you would please mention this in your model description. I'd like to see if and how this is used. It will help provide further ideas and direction for the future.

描述:

see ifap-sc main page for further details

8e-7 60epochs EMA trained against SD1.5prunedEMA

训练词语: