How accurate is AI body fat from a photo?

By ScanBodyFat. Published October 26, 2026.

Our honest margin

Every photo-based estimate carries error, and on ScanBodyFat we put the expected size of it up front: roughly 3 to 5 percentage points against a lab measurement. That is the reason the result comes back as a range with an explanation, not as a single decimal that looks more exact than a photo can be. If the tool says 20 to 24 percent, treat the real value as plausibly a little outside that band, not as a verdict.

What published research shows

Independent studies give a sense of what is possible, but they are about specific systems under specific conditions, not about this tool.

• In a study of 134 adults at two clinical sites, body fat estimated from smartphone photos by a purpose-built system had a mean absolute error of 2.16 percentage points against DXA. Three consumer scales and two professional bioimpedance systems in the same study had mean absolute errors between about 3.1 and 5.9 points. Photos were taken in form-fitting clothing from roughly 4 to 6 feet away (Majmudar et al., 2022).
• An earlier research method that used two photos taken against a green screen, plus age, sex, race and BMI, had a mean absolute error of about 4.1 percent in 323 participants, with larger errors at the lean and high ends of the distribution (Affuso et al., 2018).

Results like these show that a photo carries real information about body composition, and that the error is not zero. The reference method has error too: a published review notes that DXA results shift with food intake, hydration and prior exercise (narrative review).

What drives the error

Lighting

The authors of the 2022 validation study state that extremely dark or bright images reduce accuracy. Even, moderate light that does not hide or exaggerate shadows on your torso is the safest choice.

Pose

The same authors list holding the belly in, flexing, and large deviations from a relaxed pose as things that can hurt accuracy. Stand the way you normally stand.

Clothing

Loose or covering clothing, for example baggy shirts or shorts that cover the abdomen or thighs, can produce inaccurate results. The model can only see the outline you give it.

Camera angle and distance

Extreme camera tilt or a camera that is too far away were also listed as problems. A straight-on photo at about torso height, with your whole torso in frame, avoids both.

Body type and the extremes

Studies tend to have fewer participants at the very lean and very high ends, and error was larger there in at least one photo-based study (Affuso et al., 2018). Muscle mass, fat distribution and height also change how a given percentage looks. The height, weight, age and sex you enter help the tool account for some of this, but not all of it.

Timing

A large meal or an intense workout right before the photo can also affect accuracy, according to the same authors. Consistent timing, such as the same time of day, makes repeat photos easier to compare.

How to take a good photo

• Use even light, such as a bright room or window light from the side. Avoid harsh overhead shadows and direct flash.
• Wear fitted clothing that shows your torso outline, such as a fitted tee or athletic wear.
• Hold the camera straight at about torso height, not tilted up or down, and step back so your whole torso is in the frame.
• Stand relaxed and facing the camera. Do not flex or hold your stomach in.
• Try to avoid taking it right after a large meal or hard workout.
• For repeat photos, keep the spot, light, clothing and time of day the same.

When to use something else

If you need a clinical measurement, for example for a medical decision, ask a doctor or use a clinic method such as DEXA. If you mostly care about trends over time, a consistent method (the same scale, the same calipers measured by the same person, or the same photo setup) matters more than the absolute number. See the method comparison for the trade-offs, and the visual guide for what typical ranges mean.

Frequently asked questions

Sources

1. Majmudar et al., Smartphone camera based assessment of adiposity: a validation study, npj Digital Medicine, 2022
2. Affuso et al., A method for measuring human body composition using digital images, PLoS One, 2018
3. Come Back Skinfolds, All Is Forgiven: A narrative review of common body composition methods, PMC

More guides

• What does a body fat percentage look like?
• Photo vs DEXA, calipers and BIA: which body fat method is best?
• Can ChatGPT estimate body fat from a photo?

Estimates are for fitness purposes only, not a medical measurement or diagnosis. If thinking about body measurements feels stressful, you can skip them. For support with eating or body image concerns, see the National Eating Disorders Association.