Text-to-image generators operate with remarkable speed, yielding enduring consequences that range from the dissemination of political propaganda to the proliferation of deepfake pornography. The advancements in AI technology have significantly enhanced its ability to elude detection.

Although the industry is actively exploring watermarking and alternative approaches to identify AI-generated images, these methods lack visual cues discernible to the human eye. However, there are steps you can take to effectively evaluate images and sharpen your discernment, decreasing the likelihood that AI will deceive you. While these techniques are not infallible, they will refine your instincts for recognizing AI's involvement.

When dealing with an image of potential significance, conducting a reverse image search is crucial to ascertain its origin. It is essential to determine its authenticity rather than merely relying on its circulation on social media platforms. If extensive efforts to locate the image on reputable news websites prove futile, yet it appears groundbreaking, there is a high probability that it has been fabricated.

To illustrate this point, consider the fictional Great Cascadia earthquake of 2001. Once images depicting the non-existent event were disseminated through a subreddit, individuals who failed to conduct further investigation accepted the event as genuine. The visual portrayal of the Pacific Northwest's devastation, including collapsed roads, rescue personnel, and panicked individuals engrossed in their mobile devices, weaves a coherent narrative that can quickly go unquestioned.

It is essential to go beyond a simple reverse image search to uncover the truth. For instance, when we conducted a search using a "photo" labeled as "Amateur video captures the aftermath of the tsunami in Kodiak, AK," Google's image search yielded a snippet of a familiar-looking image, accompanied by the headline "Floods Ravage Western Canada" in the results from NBC News. At first glance, it appeared to validate the authenticity of the event. However, a single click exposed that Midjourney had illicitly utilized a photojournalist's work to fabricate a similar depiction.

As artificial intelligence constructs its creations by assembling existing elements, it may exhibit certain irregularities when scrutinized up close. To detect indications of AI involvement in an image, carefully zoom in on each component. This closer examination allows for easier identification of stray pixels, peculiar outlines, and misplaced shapes.

Consider this image portraying a bustling museum, generated using DALL-E 2. At first glance, it presents a lively cultural scene during a busy weekend day, with a crowd of people. However, upon closer inspection, you'll notice that each individual person is a composite of distinct parts derived from the AI's training dataset.

AI typically struggles to accurately replicate pores and other imperfections. If an image appears excessively flawless, it raises suspicions of its authenticity. In today's filtered online realm, distinguishing between real and altered content can be challenging. However, even amidst this context, a selfie generated by Stable Diffusion exposes itself with flawlessly retouched skin that surpasses the capabilities of Facetune.

Even Khloe Kardashian, often criticized for extensively adjusting such settings, presents a greater sense of human authenticity on Instagram. While her meticulously contoured and highlighted face approaches AI-like perfection, it still exhibits elements of light and dimension. Moreover, the texture and color variations in the skin on her neck and body further emphasize the contrast to the artificial selfie mentioned earlier.

In the example image above, showcasing a bustling street generated by DALL-E 2, a prominent clue adheres to the matching principle: the word "TAXI" painted on the street exhibits uneven spacing between its letters. Moreover, the background offers several additional hints. All the license plates are intentionally blurred, and the figure walking on the left appears indistinct in form.

Frequently, AI prioritizes the creation of the foreground in an image, leaving the background intentionally blurry or lacking clarity. Direct your attention to that blurred area, specifically searching for recognizable outlines of signs devoid of text or any peculiarities in the topographical features. While AI may be a scientific endeavor, detecting its presence requires a different kind of artistry.

In the example image above, showcasing a bustling street generated by DALL-E 2, a prominent clue adheres to the matching principle: the word "TAXI" painted on the street exhibits uneven spacing between its letters. Moreover, the background offers several additional hints. All the license plates are intentionally blurred, and the figure walking on the left appears indistinct in form.

Adhering to the principle of "It takes one to know one," utilizing AI-powered tools to identify AI creations appears to be a viable approach. Although numerous such tools are available, they often struggle to recognize their own kind. Nonetheless, they can provide valuable insights.

One such tool, AI or Not, allows you to obtain a verdict by simply dragging and dropping, uploading, or providing the URL of a questionable image. In our case, we submitted a "photo" featuring a palomino in a field, which was generated using Stable Diffusion. The outcome was unequivocal: AI.

Hive Moderation, a provider of AI-driven content moderation solutions, offers an AI detector where you can conveniently upload or drag and drop images for analysis. When we submitted the palomino image to their system, it received a remarkable score of 93.8%, indicating a high likelihood of being AI-generated.

Another reliable option is Hugging Face's AI Detector, which enables you to upload or drag and drop dubious images for evaluation. Using the identical questionable "photo," the assessment yielded a verdict of 90% human and 10% artificial, suggesting a mixed composition.