Deepfakes Explained: Risks and Realities

Deepfakes — AI-generated synthetic media that places real people's likenesses into content they didn't create or consent to — represent one of the more consequential developments in AI technology. This guide explains the technology, its application in adult content specifically, the real harms it causes, and what legal and practical recourse exists for those affected.

What Are Deepfakes?

The term "deepfake" originally referred specifically to AI face-swapping technology — placing one person's face onto another's body in video, using deep learning (hence "deep" + "fake"). The term has broadened to refer more generally to synthetic media that realistically depicts real people in scenarios they didn't participate in, using AI generation.

Deepfakes exist in several forms:

• Face-swap video: Replacing the face of a person in an existing video with another person's face — the original application of the term
• Voice cloning: AI generation of audio that replicates a real person's voice saying things they didn't say
• Fully AI-generated scenes: Text-to-video or image generation producing entirely new scenes that realistically depict real-looking people, whether based on specific individuals or not

How Deepfakes Are Created

Face-swapping deepfakes typically use one of several technical approaches:

• GAN-based face-swapping: Generative Adversarial Networks trained on many images of a target person's face, learning to replace faces in video while maintaining consistency with the target's appearance across frames
• Diffusion-based inpainting: Using diffusion models (described in our AI image generator explainer) to replace faces in existing images
• LoRA fine-tuning: Training a specialized fine-tuned model on images of a specific person to enable generation of new images featuring that person

The quality of deepfakes has improved dramatically as the underlying models have improved. Early deepfakes were detectable through obvious artifacts; more recent outputs can be very difficult to distinguish from authentic media. This technical guide specifically does not cover methods or tools for creating deepfakes — the focus here is on understanding and responding to the technology.

Deepfakes in the Adult Content Context

The majority of non-consensual intimate deepfakes target women — particularly celebrities, public figures, and increasingly private individuals. Research from organizations tracking deepfake content has consistently found that the vast majority of deepfake pornography depicts real people (typically women) without their consent.

Unlike some uses of AI-generated content (see our article on NSFW AI tools), non-consensual deepfakes of real people don't occupy any ethical gray area — they represent non-consensual intimate image abuse regardless of the AI generation method.

The Real Harms

The harms from non-consensual deepfake pornography are real and documented:

• Psychological harm: Victims report significant distress, anxiety, depression, and PTSD symptoms — comparable to other forms of sexual violence in terms of psychological impact, according to victim studies
• Reputational and professional harm: Even when viewers are told an image is fabricated, reputational damage occurs. False information about an event sticks even after correction (the "continued influence effect" in psychology)
• Harassment facilitation: Deepfake images are frequently used as part of broader harassment campaigns, including sextortion attempts
• Chilling effects: The existence of deepfake technology causes some public figures and professionals to limit their public presence or media participation to reduce their exposure as targets

Detection and Identification

Several approaches exist for detecting deepfakes:

• AI detection tools: Companies and researchers have developed classifiers trained to detect deepfake artifacts. Performance varies by tool and by the generation technique used — state-of-the-art generation can evade many current detectors
• Content provenance: The C2PA (Coalition for Content Provenance and Authenticity) standard embeds metadata in media to indicate its origin. Camera manufacturers and some platforms are beginning to implement this
• Visual artifacts: Some deepfakes show inconsistent lighting, unnatural blinking, blurry edges around face boundaries, or inconsistent skin texture — though these artifacts are becoming less common in higher-quality outputs

For a detailed look at detection capabilities, see our article can you tell AI-generated content from real?

Legal Status

The legal landscape for deepfakes is developing rapidly:

• US federal law: The DEFIANCE Act (2024) creates federal civil liability for non-consensual intimate deepfakes. The TAKE IT DOWN Act creates additional remedies. Existing CSAM law fully applies to AI-generated content.
• US state law: As of 2025, more than half of US states have enacted some form of law specifically addressing non-consensual intimate deepfakes, with criminal and/or civil penalties.
• UK: The Online Safety Act 2023 includes provisions addressing intimate image abuse including AI-generated images.
• EU: The AI Act and existing image-based abuse laws address aspects of this.

What to Do If You're Targeted

If non-consensual deepfake images of you are circulating:

1. Document: Screenshot URLs, archive pages (using archive.today), and document everywhere the content appears
2. Report to platforms: Most major platforms have reporting mechanisms for non-consensual intimate imagery — use them. Under the TAKE IT DOWN Act, platforms must respond to removal requests. StopNCII.org offers a hash-matching service that helps report across multiple platforms simultaneously
3. Legal options: Consult with an attorney about civil and criminal options under applicable state and federal law. The Cyber Civil Rights Initiative (cybercivilrights.org) maintains state law resources and can provide referrals
4. Support resources: The Cyber Civil Rights Initiative also provides crisis support for NCII victims