A new type of AI fake image, 'subtlefakes', which are less explicit but more realistic, are spreading on X and are harder to detect and regulate.

When I first saw a deepfake in 2017, the technology was crude but already shocking. Now, AI fake images have evolved, and a new form called 'subtlefakes' is spreading on X.

These images do not involve nudity or sexual acts, only slight alterations — making a red-carpet photo more revealing, making a body more exaggerated, or faking a gym selfie. Their realism is so high that even experts find it hard to tell them apart from real photos.

1. Why are they hard to detect?

Traditional deepfakes have obvious flaws, but subtlefakes only have slight changes, lacking 'contextual clues'. Actor Xochitl Gomez's parking lot photo, modified by AI, was almost indistinguishable from a real one.

2. Who profits?

Verified accounts on X are posting large numbers of these images, earning money from ad revenue sharing programs based on views. X turns a blind eye, and some images are even generated by X's own Grok.

3. Why are they harder to regulate?

Generating subtlefakes has almost zero barrier to entry — there is no need to bypass safety restrictions, and mainstream AI tools all allow generating realistic but non-explicit images of real people.

In one line: AI abuse is shifting from crude and explicit to subtle and realistic, and both platforms and regulators are falling behind.