In late 2017, an anonymous Reddit user with the handle “deepfakes” started posting videos with celebrities’ faces swapped onto other people’s bodies. The technology wasn’t entirely new. Researchers had been working on face synthesis since the 1990s, and Hollywood had spent millions doing it frame by frame. But this person had packaged the process into something anyone with a decent graphics card could run at home. Reddit banned the account and its community a few months later, but the username had already become the word.
Deepfake. A portmanteau of “deep learning” and “fake,” coined not by a researcher or a journalist, but by a pseudonymous poster on the internet. It’s one of the most loaded terms in technology, and it was named the way most slang is: casually, and by accident.
What a Deepfake Actually Is
A deepfake is any piece of media (video, audio, or image) created or altered by AI to convincingly represent something that didn’t happen. A video of a person saying words they never spoke. An image of someone in a place they never visited. A voice recording of a conversation that never took place.
The key word is “convincingly.” People have been faking photos since photography existed. Stalin was notoriously fond of erasing political enemies from group shots, and it only required an airbrush and a willingness to rewrite history. What makes deepfakes different is the scale and the accessibility. You no longer need a professional effects studio and a team of artists. Modern tools can produce convincing synthetic media on a laptop, sometimes in minutes.
How It Works
Most deepfakes rely on a technique called a generative adversarial network (GAN), invented by researcher Ian Goodfellow in 2014. The concept is elegant: two neural networks compete with each other. One, called the generator, creates synthetic images or audio. The other, called the discriminator, tries to spot the fakes. They train against each other in a loop, the generator getting better at fooling the discriminator, the discriminator getting sharper at catching fakes, until the generator produces output that’s nearly indistinguishable from the real thing.
Think of it like a writing workshop where one person writes passages imitating a famous author and another person tries to guess which passages are real and which are imitations. Over thousands of rounds, the imitator gets remarkably good.
For face-swapping specifically, the process works in three stages. First, the AI detects and isolates faces in both the source and target videos. Then it maps the source face’s features (eyes, mouth, nose, expressions) onto the target face. Finally, it adjusts lighting, color, and skin tone so the composite looks seamless. Voice deepfakes work on the same principle but with different inputs. The AI analyzes a sample of someone’s speech, extracting everything that makes their voice unique: pitch, rhythm, cadence, the specific texture of their vowels. From as little as ten to fifteen seconds of audio, modern systems can generate new speech that captures the speaker’s vocal fingerprint. This is essentially the same technology behind AI voice cloning for audiobooks, just applied with different intent.
A Blurrier Line Than You’d Think
This is the part worth sitting with. The technology behind deepfakes is not a separate, sinister invention. It’s the same generative AI technology that powers tools authors use every day.
When ElevenLabs creates an audiobook narration in a cloned voice, that’s synthetic media. When Midjourney generates a book cover from a text prompt, that’s synthetic media. When ChatGPT writes a paragraph in a specific author’s style, that’s synthetic media too. The word “deepfake” tends to get applied when the intent is deceptive or the result is used without consent, but the underlying machinery is shared.
This isn’t a reason to be afraid of the tools you’re using. It’s a reason to understand them.
Why This Matters for Your Writing Life
The technology you use shares roots with deepfakes. If you’ve generated a cover image, narrated an audiobook with a cloned voice, or used AI to create marketing content, you’ve used generative AI built on the same foundational research. That’s not a scandal. It’s the reality of how the technology developed. Understanding this connection helps you explain what you’re doing to readers who might have questions, and it makes you a more informed participant in conversations about AI and creative work.
Your voice and likeness have legal protection. As deepfake technology has become more accessible, the legal landscape has responded. The TAKE IT DOWN Act, signed into federal law in May 2025, was America’s first federal legislation directly addressing synthetic media. Forty-six states have enacted their own deepfake-related laws. For authors, this means your recorded voice (from podcast interviews, conference talks, book readings) has legal protection against unauthorized cloning. If someone creates a synthetic recording of you without permission, you have recourse.
Disclosure builds trust. When you use AI to generate or assist with content, whether that’s an AI-narrated audiobook, an AI-generated cover, or AI-assisted prose, transparent disclosure helps your readers trust you more, not less. The discomfort around deepfakes is almost entirely rooted in deception. When you’re upfront about how you use AI tools, you sidestep that problem entirely.
Detection is getting better. If you’re ever uncertain whether something you’ve encountered online is real, tools like Intel’s FakeCatcher and platforms like Reality Defender offer detection capabilities. Manual tells include unnatural blinking, misaligned lighting, and audio that doesn’t quite match lip movements. The more familiar you are with the technology, the sharper your instincts become.
The word “deepfake” carries a lot of baggage, most of it earned. But for authors working with AI tools, the most useful takeaway isn’t fear. It’s clarity. You know what the technology is, how it connects to the tools in your workflow, and where the ethical lines sit. That puts you in a much better position than vague anxiety ever could.