Cheapfakes
Also known as: Shallowfakes, Selective editing, Doctored video, Misleading edits
Low-tech video and audio manipulation (slowing a clip down, cropping, cutting, re-dubbing) that needs no AI and fools more people than deepfakes ever have.
Takes a second look
Spot it
- The clip is short and starts or stops at a suspicious moment, and the question before the answer is missing.
- Speech sounds slurred or weirdly fast while the background noise does not change.
- The audio and the picture do not quite belong together, or the subtitles say more than the speaker does.
- A longer version exists, and the people sharing the short one never link it.
- The caption tells you what the person really meant instead of letting you hear it.
How it works
A cheapfake is a lie made with the tools already on your laptop: trim, crop, slow down, speed up, swap the audio, add a subtitle. No neural networks, no training data. The researchers Britt Paris and Joan Donovan coined the term in 2019 to make a point that still holds: the fakes doing the most damage are not the high-tech ones. They are the cheap ones, because anyone can make them in minutes and they are much harder to call “fake”, because every frame is real.
The classic case is from May 2019. A video of US House Speaker Nancy Pelosi was slowed to about 75 percent of its original speed, with the pitch adjusted so that her voice still sounded natural. The result made her seem drunk or unwell. It spread across Facebook, Twitter and YouTube, was shared by prominent accounts and reached millions before fact-checkers compared it with the C-SPAN original. Nothing was added. Nothing was invented. The timeline was simply stretched.
The other common moves are even cheaper. The cut: stop the clip one second before “…which is why we reject that idea.” The crop: keep the one sign in the crowd that looks extreme and lose the two thousand that do not. The speed-up: make a shove look like an assault. The dub: keep the picture, replace the sound, or add subtitles that say what the speaker never said, since anyone who does not speak the language will trust the text. And the loop: repeat a two-second stumble until it looks like a pattern.
Why does this beat deepfakes? Because a cheapfake has a real original to hide behind. Challenge it, and the poster says: “It’s him. It’s his voice. Are you saying the video is fake?” You are not. You are saying the edit is.
Example
@accountability_now, 2h [video, 6 seconds: the minister at a press conference, laughing, cut to black] Flood victims still have no water and THIS is how he reacts to the question. Resign.
@zsofi_watches_full_clips, 1h Full press conference, 41:10. The laugh is at a journalist’s joke about the microphone; the flood question comes four minutes later and he answers it for three minutes. Not saying the answer was good. Saying this clip is not the reaction.
Notice that nothing in the six seconds is fake. The lie lives in the caption and the cut, and in what you never got to see.
Why it works on you
- Real footage disarms suspicion. You have learned to be wary of AI, so you check for melted hands and find none. A cheapfake passes every test you have been taught to run. See deepfakes for why provenance beats pixel-hunting.
- The feeling arrives first. Six seconds of a laughing minister over “flood victims” is engineered to produce anger before thought. It is emotional hijacking in a single clip.
- The caption tells you what to see. Once you have been told “this is him mocking victims”, your brain finds the mockery. The same clip captioned “minister relaxes before tough questions” would read completely differently.
What to do
When someone sends it to you: find the longer version. Search the event name plus “full” or “complete”, or look for the original broadcaster’s upload. Watch thirty seconds before and after the clip. If speech sounds odd, compare it with any other recording of the same person; slowed audio rarely matches. Treat subtitles as a claim, not a translation.
When you catch yourself: notice whether the clip is doing all the work. If you could not explain what was manipulated and why it matters without the caption, you do not know what you are sharing. Clips that cut off mid-sentence or start mid-answer deserve a pause, however good they feel. See pause before sharing.
In a comment section: post the timestamp and the full video, not an accusation. “Full clip at 41:10, the question comes later” is unanswerable. “You are spreading propaganda” starts a fight nobody wins.
Don’t confuse it with
- Ordinary editing. News reports cut every interview; what matters is whether the cut changes the meaning. A ten-second clip that fairly represents a ten-minute answer is not a cheapfake.
- Out-of-context media, where the footage is untouched and only the caption misplaces it in time or space. Cheapfakes alter the footage itself, only a little, which is enough.
Sources & further reading
- Britt Paris & Joan Donovan, Deepfakes and Cheap Fakes: The Manipulation of Audio and Visual Evidence (Data & Society, 2019)
- Washington Post: Faked Pelosi videos, slowed to make her appear drunk, spread across social media (23 May 2019)
- PolitiFact: Viral video of Nancy Pelosi speech was manipulated (24 May 2019)
- Media Manipulation Casebook: Cheap fake (definition)