You’re viewing a text-only version of this website that uses less data. View the main version of the website including all images and videos.
How to sabi foto wey dem don manipulate
In a world wia pictures fit dey digitally altered wit just a few clicks or even generated entirely from scratch by artificial intelligence (AI), e don dey become ever harder to trust wetin our eyes dey see.
Di techniques dem dey use manipulate images dey so sophisticated dat we don enta di era of di hyper-realistic fake.
Such images fit lead to di spread of misinformation and fit even influence public opinion on important events such as elections.
Wit picture of di Princess of Wales and her children wey some news agencies don begin withdraw unto concerns say e dey "manipulated", di issue don bring plenty attention to di issue like never before.
So, e get anytin we fit do to spot an image wey dem fit don alter or one wey dey AI generated?
Reflections and shadows
Abnormal lighting na often a giveaway sign say a photograph dey altered.
Check di points of light for pipo eyes, for example; di source of light go often dey reflected in dia eyes.
If di size and colour no match wit di location, or if dem look different between di eyes, den you fit get cause to dey suspicious. Di way subjects and objects dey appear for reflective surfaces for any image fit also provide clues.
End of Di one wey oda users dey read well well
Shadows of objects in di image fit no line up if dem join dem togeda from multiple pictures, although bear in mind say some pictures dey taken with multiple lighting sources.
E dey important to look di way light dey play across a subject face. If di sun dey behind dem, for example, dia ears go look red.
Artificial intelligence fit also produce incongruent lighting and shadows, but as algorithms improve, AI-generated faces often dey believe as if e dey more real dan human faces.
Hands and ears
Anoda tell-tale approach na to look for features wey dey hard to reproduce.
AI currently dey poor in rendering hands and ears, as e dey spoil dia shapes, proportions and even di number of fingers.
Dis na di same features wey dey often trouble artists, but as oda aspects of AI-generated images of pipo become hyper-real, these inaccuracies create one kain extraodinary valley wey look unnatural to our eyes.
Look di metadata
Sometin wey dem hide within di code of digital images na nuggets of information wey fit help to pinpoint a fake.
Evri time a digital camera snap image, metadata dey written into di image file.
Timestamps, for example, na im lead to questions about weda President Donald Trump in fact dey work in di White House di day afta e reveal say e don catch Covid-19 in October 2020.
Image noise
Evri digital camera sensor dey feature tiny manufacturing faults wey fit lead to unique errors wey dey leave a kind of "fingerprint" on pictures.
Dis one dey associated with a specific camera and e fit help to identify areas of a photograph wey dey manipulated. Di grain of an AI-generated image fit also look peculiar.
Verification tools
0Internet companies such as Google don launch image verification tools wey fit help pipo to spot AI-generated images.
Facebook and Instagram don start to label AI-generated images wey come from Meta dia own systems and dem dey plan to do di same for pictures wey dey generated by of companies AI tools.
