Disclaimer: This tool is for personal informational use only and should not be used for medical diagnosis or treatment decisions. For health-related concerns, please consult with a qualified healthcare professional.
These Results Are a Floor, Not a Ceiling
This validation used minimal inputs: a single front-facing photo with age and weight. The API also accepts:
- Side photo for better fat distribution analysis, especially around the midsection and lower back
- Height for more accurate proportion and body volume estimates
- Previous results so the model can calibrate to your individual body over time
None of these were available in our test data.
Users who provide additional inputs should see even better accuracy than what we report here.
94% Accuracy in the Most Common Body Fat Range
For users between 15% and 25% body fat, 94% of our estimates landed within ±5% of their DEXA-verified result. The average error in this range was just 2.4%.
This is the range most people care about. If you're cutting, lean bulking, or recomping, the model performs well exactly where you need it to.
3.1% Average Error Overall
Across all body fat levels tested, from under 7% to nearly 40%, our API achieved a mean absolute error of 3.1% against DEXA.
| Method | Typical Error vs DEXA | Cost | Convenience |
|---|---|---|---|
| Our API | ~3.1% | $0.10 | Phone photo |
| Skinfold Calipers | 3-4% | $10-30 | Requires trained technician |
| Bioimpedance (BIA) Scales | 3-5% | $30-200 | Affected by hydration, meals |
| Bod Pod | 2-3% | $50-100 | Requires clinic visit |
| DEXA Scan | Gold standard | $75-150 | Requires clinic visit |
That puts us right alongside methods that need equipment, trained operators, or clinic visits.
82% of Estimates Fall Within DEXA's Own Error Margin
DEXA is treated as ground truth, but it has real limitations. Research published by Weightology shows that DEXA individual error rates typically sit around 4-5%, with some studies reporting 8-10% depending on the machine and calibration.
82% of our predictions fell within that ±5% window. For most users, our estimate lands inside the range a DEXA scan itself might produce.
More Consistent Than DEXA
We ran the same photos through our model multiple times. The average difference between runs was ±0.4%.
DEXA results shift based on hydration, meal timing, recent exercise, body positioning on the table, and which machine you use. A study on DEXA precision found ±0.8% variance for 68% of repeat scans under controlled conditions.
Our model beats that. A photo doesn't change between measurements.
Near-Zero Bias
The average signed error across all tests was -0.7%. The model doesn't consistently read high or low. The overall average prediction was within 0.7 percentage points of the overall DEXA average.
Equally Accurate for Men and Women
We validated across male and female subjects and found nearly identical results:
The model handles different body types and fat distribution patterns without needing separate calibration.
Methodology
We collected data from real people who had undergone DEXA or DXA scans and shared their results alongside photos of their physique. Each entry was verified to ensure:
- The photo showed a single person with a visible, minimally clothed torso
- The DEXA/DXA result came from an actual scan
- No text overlays, composite images, or before/after collages were present
- The photo and scan result were from the same time period
Photos were submitted to our API with age and weight. No special preprocessing, no cherry-picking.
Entries where the person indicated their DEXA result was anomalous or disputed were excluded to avoid skewing results with known measurement errors.
Precision testing was done by submitting the same photos on separate occasions and measuring variance between results.
A comparable result for 99.9% lower cost
A DEXA scan costs $75-150, requires a clinic visit, and carries a 4-5% individual error rate. Our API costs $0.10, works from a phone photo, and delivers comparable accuracy.
For tracking progress, setting goals, and understanding your body composition, this is the most practical way to get a reliable number.