An AI-generated image is any picture made by AI, from a dragon in space to a fake product photo. A deepfake is AI-made or AI-altered media that looks like a real person, place or event and could fool you into thinking it is genuine. So every deepfake picture is AI-made or AI-edited, but most AI images are not deepfakes. Deepfakes can also be video or a cloned voice.
| Question | AI-generated image | Deepfake |
|---|---|---|
| What it is | Any picture an AI tool made or largely made. | AI-made or AI-altered media that resembles something real and could pass as genuine. |
| Copies someone or something real? | Not necessarily. It can be pure fantasy. | Yes: a real person, place, object or event. |
| Could fool you? | Not necessarily. A cartoon fools no one. | Yes. That is what makes it a deepfake. |
| Formats | Still pictures. | Pictures, video and audio. |
| Starts from | Usually a text prompt. | Often real photos, footage or voice clips of the person. |
| EU rule (AI Act) | The company whose tool made it must mark it as AI-made in a machine-readable way. | The business or person using the tool for work must also say it is AI-made. |
The table is this site’s summary of the sources quoted below. The EU rules are law in the European Union only.
Think of “AI-generated image” as the word “drawing” and “deepfake” as the word “forgery”. Every forgery is a drawing of some kind, but most drawings are not forgeries. What turns one into the other is copying something real and passing it off as genuine.
What is an AI-generated image?
The plain point: any picture an AI tool made, whatever it shows. A text-to-image tool that turns a sentence into a picture is the usual way.
The EU’s AI law, the AI Act, treats all of it the same way at the point of making. Companies whose tools generate “synthetic audio, image, video or text content” must make sure the results are “marked in a machine-readable format and detectable as artificially generated or manipulated.”1 That mark can be a hidden watermark or a signed record such as Content Credentials.
Most AI images are posters, illustrations and fantasy scenes. Being AI-made does not make a picture a deepfake.
What makes something a deepfake?
The plain point: two things together. It looks like something real, and it could be taken as genuine.
The EU AI Act gives the clearest legal definition: “‘deep fake’ means AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful”.2 Note the words “generated or manipulated”: a real photo with a face swapped in counts, not only a picture made from scratch.
The word itself was coined in 2017, joining “deep” from deep learning with “fake”.3 In everyday talk it usually means a real person, faked. See deepfake meaning.
Ask two questions: does it copy someone or something real, and could a reasonable person believe it? Two yeses make a deepfake. Anything else is just an AI image.
Which is which? Five examples
The plain point: the same tool can make either; what matters is what the picture shows and how it is used. These calls are this site’s reading of the EU definition above.
- A cat in a spacesuit on the Moon. AI-generated image. No real event, and it is plainly made up.
- A made-up face for a game character. AI-generated image. It looks real, but it copies no real person.
- A realistic photo of a real politician at a protest they never attended. Deepfake. A real person, and it could pass as a news photo.
- A real photo with a stranger removed by inpainting. An AI edit. Whether it misleads depends on what was changed and how it is used; removing someone from a news photo is a different matter from tidying a holiday snap.
- A clearly labelled satire video of a real politician. Still a deepfake by the definition, but the EU’s labelling duty is lighter for obvious satire, art and fiction (see the next section).
What do the rules say about each?
The plain point: in the EU, every AI image should carry a machine-readable mark from the tool, and a deepfake also needs a plain disclosure from the business or professional who makes and uses it.
Article 50 of the AI Act splits the duty. Providers, meaning the companies whose tools make the content, must mark all of it (quoted above). Deployers, meaning the people or organisations using the tool, must disclose deepfakes. The law’s definition of a deployer leaves out use “in the course of a personal non-professional activity”, so this duty is aimed at work use, not private fun.4 Deployers they “shall disclose that the content has been artificially generated or manipulated.” For evidently artistic, creative, satirical or fictional work, that disclosure can be done in a way that does not spoil the work.5 Dates and the rules outside the EU are on how AI content gets labelled.
Other laws can apply too, such as fraud, harassment or the right to your own face and voice. Those differ by country and, in the US, by state; cloning a voice or a face explains the US position.
Why are deepfakes the bigger risk?
The plain point: because they are built to be believed, and they are used to get money or trust out of people.
The US Federal Trade Commission warned in 2023 that a scammer needs only “a short audio clip of your family member’s voice” and a voice-cloning program to fake an emergency call.6 In 2024 an employee at the engineering firm Arup joined a video call where the senior colleagues he saw were “deepfake re-creations”, and sent the money they asked for.7 An AI picture of a cat in a spacesuit does none of that.
Can you tell them apart?
The plain point: often not by looking. Check where the picture came from instead.
- Detectors are weak on real fakes. A 2025 study of deepfakes that actually circulated in 2024 found open-source detectors scored far worse than on older tests: “AUC decreasing by 50% for video, 48% for audio, and 45% for image models compared to previous benchmarks.”8 AUC is a common score for how well a detector separates real from fake.
- Labels help when present, but can be stripped. Content Credentials can be separated from a file, C2PA says.9 NIST, the US standards body, says “most current watermarking schemes have been consistently found vulnerable to removal”.10 A missing label proves nothing.
- Find the earliest copy. A reverse image search shows where else a picture appears. In Google, you “Click Search by image and then Upload a file”.11
You do not need to know whether a picture is AI-made to protect yourself. If it shows a real person doing something surprising, check the source before you believe or share it.
Questions people ask
Is every AI image a deepfake?
No. A deepfake must resemble a real person, place, object or event and be able to pass as genuine, under the EU definition quoted above. A fantasy picture or a made-up face is an AI image, not a deepfake.
Is a deepfake always made from scratch?
No. The EU definition covers content that is “AI-generated or manipulated” (source above), so a real photo or video with a face swapped or words changed counts too. Many deepfakes start from real footage of the person.
Is a cloned voice a deepfake?
Yes, if it imitates a real person and could pass as genuine. The EU definition names image, audio and video (source above). See voice cloning meaning.
Where to go next
- “Is this picture real?” Use the spot AI fakes checklist, then how to spot fake images, video and voices.
- “Someone is calling in a panic.” Read deepfake self-defence and how AI fraud works.
- “Do detectors work?” See do AI detectors work?
- “How are AI images labelled?” Read how AI content gets labelled, Content Credentials and AI watermarking.
- “I want to animate a real person’s photo.” Do it with consent: how to make a talking photo with AI.
- “What do the words mean?” See deepfake, text-to-image and AI slop.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What two things make something a deepfake, according to the page?
- It is made with AI and posted online
- It resembles something real and could pass as genuine
- It is a video and has sound
Show the answer
Answer: It resembles something real and could pass as genuine. The page uses the EU definition: content that resembles real persons, places or events and would falsely appear authentic.
2. Is every AI-generated image a deepfake?
- No, a fantasy picture or made-up face is just an AI image
- Yes, any AI picture counts as a deepfake
- Only if it was made on a phone
Show the answer
Answer: No, a fantasy picture or made-up face is just an AI image. The page says most AI images are not deepfakes, because they do not copy something real and pass it off as genuine.
3. What does the page say is the best way to check a surprising picture of a real person?
- Trust any AI detector score
- Assume it is real if it has no label
- Find where it came from, for example with a reverse image search
Show the answer
Answer: Find where it came from, for example with a reverse image search. The page says detectors are weak and labels can be stripped, so check the source and the earliest copy.
- 1EU AI Act, Article 50(2), via artificialintelligenceact.eu, Article 50, read at source 19 Oct 2026providers of AI systems “generating synthetic audio, image, video or text content,” “shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated.” ↑
- 2EU AI Act, Article 3(60), via artificialintelligenceact.eu, Article 3: Definitions, Regulation (EU) 2024/1689, read at source 1 Oct 2026 (as cited on this site’s deepfake meaning page): “‘deep fake’ means AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful”.↑
- 3Dictionary.com, deepfake, dictionary entry, read at source 12 Oct 2026 (as cited on this site’s deepfake meaning page): “First recorded in 2015–20; deep (learning) + fake”; “Deepfake was coined in 2017.”↑
- 4EU AI Act, Article 3(4), via artificialintelligenceact.eu, Article 3: Definitions, read at source 19 Oct 2026“‘deployer’ means a natural or legal person, public authority, agency or other body using an AI system under its authority” … “except where the AI system is used in the course of a personal non-professional activity”. ↑
- 5EU AI Act, Article 50(4), via artificialintelligenceact.eu, Article 50, read at source 19 Oct 2026deployers of an AI system “that generates or manipulates image, audio or video content” constituting a deep fake “shall disclose that the content has been artificially generated or manipulated.” Summary of the same paragraph: for evidently artistic, creative, satirical or fictional work, the duty is limited to disclosing that such content exists in a way that does not hamper the display or enjoyment of the work. ↑
- 6US Federal Trade Commission, Scammers use AI to enhance their family emergency schemes, consumer alert, 20 Mar 2023, read at source 1 Oct 2026 (as cited on this site’s deepfake meaning page): “All he needs is a short audio clip of your family member’s voice — which he could get from content posted online — and a voice-cloning program.”↑
- 7CNN Business, report on the Arup deepfake call, 16 May 2024, read at source 17 Sep 2026 (as cited on this site’s deepfake meaning page): the employee met people “he believed were the chief financial officer and other members of staff, but all of whom turned out to be deepfake re-creations.”↑
- 8Chandra et al., Deepfake-Eval-2024: A Multi-Modal In-the-Wild Benchmark of Deepfakes Circulated in 2024, arXiv, 4 Mar 2025, read at source 5 Oct 2026 (as cited on this site’s AI detectors page): “AUC decreasing by 50% for video, 48% for audio, and 45% for image models compared to previous benchmarks.”↑
- 9C2PA, FAQs, read at source 1 Oct 2026 (as cited on this site’s Content Credentials meaning page): “While C2PA Manifests are typically embedded in the asset, they can be separated.”↑
- 10NIST, NIST AI 100-4, November 2024, read at source 1 Oct 2026 (as cited on this site’s watermarking meaning page): “most current watermarking schemes have been consistently found vulnerable to removal”.↑
- 11Google Search Help, Search with an image on Google, read at source 5 Oct 2026 (as cited on this site’s spot AI fakes checklist): “Click Search by image and then Upload a file”.↑
Read at source 19 Oct 2026: EU AI Act (Regulation (EU) 2024/1689), Article 50 and Article 3(4), via artificialintelligenceact.eu.
Reused from this site’s deepfake, Content Credentials, watermarking, AI detectors and spot AI fakes pages, with the read dates recorded there: EU AI Act Article 3(60), Dictionary.com, the US Federal Trade Commission, CNN Business, C2PA, NIST AI 100-4, Deepfake-Eval-2024 and Google Search Help.
The comparison table, the drawing-and-forgery comparison, the five example calls and the answers marked as this site’s reading are this site’s own, not claims from these sources, and not legal advice.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK