How it works
What a photo can actually tell you about your body fat.
Computer vision models can estimate body fat percentage from a single photo by reading visual markers, muscle definition, subcutaneous fat distribution, body shape, and mapping them to an estimate. It's not magic, and it's not clinical. Here's what's actually happening, what it's worth, and where the limits are.
How photo-based estimation actually works
A model is trained by observing large numbers of photo pairs and reference measurements (typically DEXA or water-displacement methods), learning which visual markers, definition, shading over muscle groups, overall shape, correspond to which body-fat range. Applied to a new photo, the model produces an estimate based on those learned patterns.
Quality depends entirely on capture: consistent framing, distance, pose, and lighting reduce noise. A poorly framed, badly lit, or baggy-clothing photo confuses the model the same way it would confuse a trained human eye.
See Corpus's full methodology →How this compares to other methods
| Method | What it measures | Cost and access | Repeatable at home |
|---|---|---|---|
| DEXA scan | Bone mineral density and tissue composition, via dual-energy X-ray | $50–150 per scan, requires a clinic | No, 1–2x per year at most |
| Skinfold calipers | Subcutaneous fat thickness at specific sites | Low cost, requires trained technique | Yes, but high variance between measurers |
| Bioelectrical impedance (smart scale) | Electrical resistance through the body, converted to fat % | $30–150 scale | Yes, but sensitive to hydration and timing |
| Photo-based estimate (Corpus) | Visual markers of definition and fat distribution | Included with subscription | Yes, with consistent capture |
The part most AI apps don't tell you
No home method, calipers, a smart scale, a photo, is going to match DEXA-scan precision. That's not a Corpus-specific limitation; it's true of any estimate made outside a lab. What actually differs between methods is consistency: whether the method gives you the same answer under the same conditions, so a change in the number reflects a real change in you, not just measurement noise.
Corpus doesn't claim to replace a DEXA scan, and isn't trying to. It exists for the other 363 days of the year, when a lab visit isn't realistic but you still want a real signal, read the same way every week.
Accuracy questions
How accurate is a photo-based body fat estimate?
Photo-based estimates will not match a DEXA scan to the decimal, and Corpus does not claim they will. Consistent capture conditions can make the trend more useful over time, but every result remains an estimate rather than a clinical measurement.
Is this a replacement for a DEXA scan?
No, and it isn't trying to be. A DEXA scan is a clinical-grade snapshot you take once or twice a year. Corpus is built for the other 363 days, when a lab visit isn't realistic but you still want a real signal.
What happens to my photos?
A check-in photo is sent for analysis and handled as described in the Corpus Privacy Policy. It is not sold, used for advertising, or placed in a public photo library. Review the policy for current processing, storage, and deletion details.
Start a guided photo check-in
Consistent capture, the same analysis every time, and a trend you can actually follow.