Live here: Bias detection in SinteticoXL Models
This repo is a mirror from the Space originally published on Hugging Face.
In order to enhance model transparency and ensure ethical AI practices, I made a fork of text-to-image-bias to apply their approach to my own merged models.
In this demo, we explore the potential biases in text-to-image models by generating multiple images based on user prompts and analyzing the gender, skin tone, and age of the generated subjects as well as the potential for NSFW content. Here's how the analysis works:
- Image Generation: For each prompt, 10 images are generated using the selected model.
- Gender Detection: The BLIP caption generator is used to elicit gender markers by identifying words like "man," "boy," "woman," and "girl" in the captions.
- Skin Tone Classification: The skin-tone-classifier library is used to extract the skin tones of the generated subjects.
- Age Detection: The Faces Age Detection model is used to identify the age of the generated subjects.
- NSFW Detection: The Falconsai/nsfw_image_detection model is used to identify whether the generated images are NSFW (not safe for work).
- Sintetico XL: a merged model with my favorite aesthetics
- Sintetico XL Prude: a SFW version that aims to remove unwanted nudity and sexual content.
Feel free to contribute!