Copyright © 2025 All Rights Reserved
5 Min
Nowadays, anyone can create the images and videos they desire with the help of artificial intelligence. It’s undoubtedly a technology accessible to all. There are already many AI-generated personas; for instance, the Spanish agency Clueless created Aitana, their beautiful virtual model, which receives thousands of messages and has caught the attention of footballers and celebrities.
While in this case, she was harmless, what if someone were to use AI for scams, impersonating famous individuals or professionals, like doctors or consultants? In order to steer clear of these pitfalls, this article will delve into methods for identifying AI-generated images and videos.
Deepfakes are videos, audio, or images manipulated using advanced AI algorithms to create fake content. Moreover, this technology enables the replacement of a person’s face in a video with someone else’s, manipulation of speeches, or fabrication of entirely fictitious situations. Generally, a deepfake can make it seem like someone is saying or doing something they have never actually said or done.
Deep learning algorithms analyze vast amounts of data, such as videos and photos of the person to be imitated, to convincingly replicate their facial movements, expressions, and vocal tones. Consequently, once the model has been trained, it can be used to generate fake content with increasing accuracy.
Furthermore, well-known applications like DALL-E by OpenAI and Midjourney allow text-to-image generation. These tools are trained by their respective companies using vast amounts of images, enabling them to “learn” how to reproduce non-existent human beings, common objects, realistic environments, and more.
The advancement of AI technology makes it increasingly difficult to differentiate between real and AI-generated images. Nevertheless, certain details can help discern whether it is a deepfake or not. These details include errors in the representation of anatomical parts, such as fingers (along with hair and skin reproduction in portrayed characters), errors in object replication, elements out of context in the image, such as people in the background with strange, unnatural features, even with twisted arms or legs.
Currently, AI might not be flawless. Frequently, it produces images containing errors – some more noticeable than others – that can offer clues to distinguish whether a piece is AI-generated. Here are 10 useful details to recognize AI-generated images:
AI algorithms still struggle to generate accurate hand representations. Images containing human characters may be identifiable due to hands with six or more fingers, missing thumbs, or fingers merging.
AI sometimes fails to replicate teeth accurately, neglecting distinctions between incisors, canines, and wisdom teeth. When dental arches appear unnaturally regular, the image is likely artificial.
In some pictures, the faces of portrayed characters may appear distorted or asymmetrical, serving as a telltale sign of artificiality.
AI doesn’t always convincingly replicate hair. Inconsistencies in hair texture, ranging from variegated and unnatural to extreme cases where hair transforms into an accessory like a scarf, are indicators of AI generation.
Texture issues extend beyond hair to skin. When the skin lacks the typical imperfections and details of human skin, and characters resemble renders, this is another detail to consider.
Images may contain architectural flaws like misaligned steps, oddly curved walls, or inexplicable slanted ceilings, typical of AI-generated images.
Deformed jewelry or oddly elongated or suspended objects held by depicted subjects are characteristic of artificially generated images.
Images with artifacts, such as missing limbs, three-eyed characters, or graphically inconsistent elements, likely result from AI.
Texts within elements of AI-generated images (e.g., on posters, street signs etc.) may exhibit artifacts, giving them an unnatural appearance.
Algorithms generating images may attempt to hide errors by blurring critical areas like the background. Alternatively, users may intentionally crop photos to conceal errors (e.g., those related to hand fingers).
Regarding deepfake videos, they are manipulations created using AI. Despite being well-made, they may still need to be corrected. Let’s see:
One common flaw is lip synchronization. In some cases, lip movements may not perfectly match the audio, resulting in an unnatural appearance.
Similar to images, deepfake videos may contain blurriness or distorted pixels, especially in areas where modifications occur.
Details like facial lighting, eye reflections, or facial expressions may be challenging to replicate convincingly, leading to unrealistic or unnatural movements.
A deepfake created from different video fragments may lack temporal coherence in actions and movements, resulting in visual incongruences or background errors.
If the deepfake involves audio manipulation, there may be a loss of audio quality.
When identifying a deepfake, attention to detail is essential. For instance, if a person’s face isn’t perfectly frontal, it may flicker, especially when simple hand movements pass in front of it.
It’s important to recognize that only some have the expertise to distinguish AI-generated images. Furthermore, the software used to generate images is constantly improving, making the task even more challenging. Therefore, it could be helpful to use specific online tools that allow users to scrutinize images suspected of being AI-generated. Among the best available tools, we have Deepware Scanner, FotoForensic, Intel’s Real-Time Deepfake Detector and Illuminarty. Using them is quite simple: just upload the image suspected to be artificial or paste its URL into their respective text field, initiate the examination process, and await the algorithm’s verdict.