Stable Diffusion is an open-source latent diffusion model released in 2022 by Stability AI. Unlike proprietary models with built-in safety filters, Stable Diffusion's open-source nature means it can be run locally without content restrictions, making it the primary tool for generating non-consensual intimate imagery. Perpetrators use Stable Diffusion with custom LORA models (fine-tuned on a victim's photos) to generate realistic intimate imagery of specific individuals. The implications of this technology are significant and warrant attention.

Key facts about this term

  1. Understand that Stable Diffusion output is significant Any AI-generated intimate content presents risks regardless of the generation tool used.
  2. Biometric scanning identifies Stable Diffusion output ScanErase's facial geometry matching identifies deepfake content including Stable Diffusion-generated imagery.
  3. Report issues with Stable Diffusion output to model repositories Custom LORA models on HuggingFace and CivitAI that generate NCII can be flagged to those platforms.

Frequently asked questions

Can Stable Diffusion's output be controlled?

Stable Diffusion's open-source nature makes it difficult to prevent local use without safety filters. However, all output content raises concerns, and those creating it may face consequences.

What is a LORA model in the context of Stable Diffusion?

A LORA (Low-Rank Adaptation) is a fine-tuned model trained on specific images, for example, photos of a specific person, to generate imagery of that person. LORA models targeted at specific individuals are used to create personalized deepfake NCII.

Related glossary terms