Anyone with an internet connection and access to an artificial intelligence (AI) tool can create photorealistic images within seconds, for free or for very little money, and then spread them on social networks at breakneck speed.
asserted
Anyone → create → speed
The tools used to create AI-generated images are improving rapidly, reducing the number of obvious AI errors – often referred to as AI hallucinations – and making it increasingly difficult for users to determine whether a viral image depicts a real event.
asserted
image → use → event
Gone are the days when AI tools routinely produced images of public figures with four fingers, an excessive amount of teeth or unnatural body proportions.
asserted
tools → go → teeth
An additional challenge is that while many images are entirely AI-generated, others are real photos that have been partially altered using AI.
asserted
that → generate → AI
Here's one example: an authentic broadcast image from the 2026 World Cup in which a fan was replaced by an Adolf Hitler lookalike.
asserted
fan → replace → lookalike
Of course, the image was actually created using an OpenAI-powered tool.
asserted
image → create → tool
So how can you spot AI-generated images?
asserted
you → spot → images
DW Fact check offers tips to help you navigate today's information landscape.
asserted
you → offer → landscape
Look closely at the picture
You may remember the story of a Syrian teenager who helped stop a knife attack in the German city of Hamburg in 2025.
uncertain
who → look → 2025
As the story spread, conflicting accounts of the incident emerged on social media, along with AI-generated images like the one above.
asserted
accounts → spread → one
The image looks authentic at first glance, but if you zoom in and examine it carefully, you'll find obvious inconsistencies.
asserted
you → look → inconsistencies
That's why it's important to take your time and inspect the details.
asserted
it → take → details
Common AI errors might be hidden in the background, including distorted objects and gibberish-filled text.
uncertain
errors → hide → objects
If your device doesn't allow you to zoom in far enough, online tools can help.
asserted
tools → allow → ?
A tool we often use at DW Fact check is Forensically, a free online photo forensics platform that includes a magnifier and other features for examining images closely.
asserted
that → use → images
A supporter resembling the Nazi dictator would almost certainly have attracted immediate attention from broadcasters, stadium officials and other spectators.
asserted
supporter → resemble → broadcasters
Yet no reputable media organizations reported such a stir, and no fans in the stadium shared posts showing the lookalike from a different perspective.
asserted
fans → report → perspective
This example highlights a key verification technique: move beyond the pixels and ask yourself whether the event itself makes sense.
asserted
event → highlight → sense
One way of doing that is checking whether the apparent event has been covered by trustworthy news organizations.
asserted
event → do → organizations
If an image appears to show a highly unusual or controversial incident, but there is no evidence of it in credible reporting, that should raise a red flag.
asserted
that → appear → flag
You can also look at who's posting and sharing the image.
asserted
who → look → image
If the account often shares AI-generated material or appears to have no connection to the event itself, it's wise to treat the image with caution.
asserted
it → share → caution
Find the image source
If you're unsure whether an image is real or generated by AI, try to find its source.
asserted
image → find → source
One way to do this is by conducting a so-called reverse image search.
asserted
way → do → search
Instead of searching with keywords, you can upload an image to tools like Google Imagesor TinEye.
asserted
you → search → TinEye
These tools can help answer several questions:
- When was the image first uploaded?
- On which platforms is the image being shared?
- What are users saying about it in comments and posts?
- Have any reputable fact-checkers already verified the image?
asserted
checkers → help → image
Of course, some AI-generated images may be brand new and therefore produce few or no results.
But manipulated images are often based on older, authentic pictures.
uncertain
images → generate → pictures
In those cases, a reverse search can reveal the original image and provide important context.
asserted
search → reveal → context
If you find that an image has already been fact-checked, one way to assess the credibility of the article or video is to see whether the organization is certified by the International Fact Checking Network (IFCN).
asserted
organization → find → Network
The IFCN assesses fact-checkers against a set of professional standards, which include being as fair and transparent as possible, being non-partisan, and committing to open and honest corrections.
asserted
which → assess → corrections
You can find a list of IFCN-certified outlets here, including DW Fact check.
asserted
You → find → check
The IFCN should not be confused with the Global Fact-Checking Network (GFCN), a Russian platform with close ties to the Kremlin.
asserted
IFCN → confuse → Kremlin
Use AI tools with caution
asserted
Use → use → caution
AI detectors can help assess whether an image was created or modified using AI.
asserted
image → help → AI
They search for invisible watermarks and other digital traces associated with AI-generated content.
asserted
They → search → content
Tech companies such as Google and OpenAI, the creator of ChatGPT, have developed systems that identify whether their own AI models were used to create or modify an image.
asserted
models → develop → image
Google's SynthID tool can be accessed here, and hereis the link to OpenAI's verification tool.
asserted
link → access → tool
Other tools, such as Image Whisperer, AI or Not, and Hive Moderation, estimate the likelihood that an image was manipulated using AI.
asserted
image → estimate → AI
And then AI chatbots, such as X's Grok, can provide answers to user questions on whether images are real or fake.
asserted
images → provide → questions
In the Hitler lookalike example, OpenAI identified a digital watermark, which confirmed it was manipulated using the AI company's tools.
asserted
it → identify → tools
…and 10 more, not listed.
The Aporia
