No software required, no guarantee either: manual deepfake detection means looking for a small set of visual cues that AI image and video generation still tends to get wrong. It catches a real share of fakes, particularly lower-effort ones, but it is a heuristic, not a verdict.
This is a practical checklist of what to actually look for, followed by an honest account of where eyeballing a photo stops being enough.
- Five visual cues to check: hands and fingers, lighting and shadow consistency, skin texture, background artifacts, and consistency across multiple photos of the same claimed moment.
- These are honest, imperfect heuristics — none of them proves a photo is fake on its own, and high-quality generations increasingly avoid the obvious mistakes.
- For a harder case, or a public figure's photo you can't resolve by eye, dedicated detection tools do a more thorough job than visual inspection alone.
The five-point checklist
1. Hands and fingers. This is the most commonly cited tell because generative models historically struggled with the exact number and articulation of fingers. Look for extra or missing digits, fingers that bend at an implausible angle, or hands that look slightly wrong without you being able to say exactly why.
2. Lighting and shadow consistency. Check whether the direction and color of light on the face matches the direction and color of light in the background. A face lit from the left in a scene where every shadow falls to the left of every other object is consistent; a mismatch is a flag.
3. Skin texture. Real skin has pores, fine lines, and small imperfections. AI-generated skin sometimes reads as unnaturally smooth or slightly waxy, especially in close-up. This cue is getting harder to rely on as models improve, but it's still worth a glance.
4. Background artifacts. Look at straight lines (door frames, window edges, text on signs) for subtle warping, and check for objects that repeat, blur strangely, or don't quite make physical sense. Backgrounds get less generation attention than faces, so errors tend to show up there first.
5. Consistency across multiple photos. If several photos claim to document the same event or moment, check whether the background, lighting, and small details (clothing, jewelry, the position of objects) actually match across all of them. A single suspicious photo is harder to judge than a set that contradicts itself.
For a public-figure photo you still can't resolve, see the best tools to detect deepfake celebrity photos.