| This is the SF_EB_1.1 model, trained by the SmirkingFace project (https://patreon.com/smirkingface and https://github.com/smirkingface/stable-diffusion). When redistributing this file, include this readme.txt file and the license.txt. By using this model you are agreeing to the included license agreement. |
|
|
| This model was trained using the original Stable Diffusion architecture, and can be run as a drop-in replacement for any Stable Diffusion model or merged with other models. It already includes the improved MSE VAE from https://huggingface.co/stabilityai/sd-vae-ft-mse-original. |
|
|
| Note that the captions used to train model were quite specific (see dataset_captions.txt), with any model names replaced with the generic 'woman'. The prompts you would normally use with Stable Diffusion may not work as well with this model. See example_captions.txt for some examples that are known to produce good results often. Also see https://www.patreon.com/posts/getting-started-73775030 for basic examples, and see https://www.patreon.com/posts/word-lists-73913464 for word frequency lists. |
|
|
| Note that this finetuned model is especially responsive to the following words (and some synonyms of them): posing, masturbating, embracing, kissing, cunnilingus, blowjob, sex, or lesbian sex |
|
|
| Training details: |
| Resumed from: SF_EB_1.0 |
| Dataset: 5.8k high quality erotic and pornographic images |
| Tagged with: |
| - The image subject: posing, masturbating, embracing, kissing, cunnilingus, blowjob, sex, or lesbian sex |
| - Number and gender of persons: one woman, one woman and one man, or two women |
| - Clothing: clothed, partially clothed, or naked |
| Data augmentation: Random rescaling, cropping, and training using padded non-square images at resolutions 384x512, 512x384, 512x512, 512x768, and 768x512 |
| Epochs: ~780 at a base learning rate of 1.0e-6 |
