Can ChatGPT estimate body fat from a photo?
By ScanBodyFat. Published .
Short answer
A general chatbot such as ChatGPT can look at a photo and give a number, but nothing checks that number against a measured result, and answers can vary from one try to the next. A structured tool that combines a photo with your height, weight, age and sex, and returns a range with its margin stated, is a more dependable way to get an estimate.
What a general chatbot can and cannot do
Chatbots that accept images can describe what they see, and many will give a body fat number if you ask. The question is whether that number is trustworthy. A few limits are worth knowing before you rely on one.
- No calibration against measured body fat. A purpose-built system is trained and tested against a reference such as DXA, and the best studies report the error (for example, 2.16 percentage points in a study of one such system, Majmudar et al., 2022). A general chatbot is not designed or validated for this task, and we did not find a peer-reviewed study that tests it.
- Limited context. Height, weight, age and sex change how a body shape maps to a percentage. A chatbot only has them if you type them in, and nothing makes you do it the same way each time.
- Inconsistent answers. Generative models can answer differently when the wording, the photo crop or the conversation history changes. If you ask twice, you may get two numbers, with no stated margin of error.
- Unknown error for similar tasks. In a peer-reviewed test on food photos, a different visual estimation task, general models including ChatGPT-4o showed mean absolute percentage errors around 36 percent for portion weight, and the authors concluded they were not yet suitable for precise dietary assessment (study). That does not prove anything about body fat, but it shows how large errors can be when general models estimate quantities from pictures.
Why a structured tool can be better
A tool built for one job can ask for the same inputs every time (a photo plus sex, age, height and weight), combine them in a consistent way, and report a range rather than false precision. ScanBodyFat works this way and states its margin plainly, roughly 3 to 5 percentage points. That does not make it a lab test. It makes the output easier to compare over time and easier to interpret. Our accuracy guide explains what drives the error and how to take a good photo.
Privacy considerations
- A photo of your body is sensitive. Before uploading it anywhere, check how long it is kept and whether it can be used for training.
- Chat history may be saved to your account by default in many chat services, so look at the settings.
- On ScanBodyFat, the photo is used only to produce your estimate. How it is processed and how long anything is kept is described in the privacy policy.
Other options
If you want a more precise number, a clinic method is the better route. See the comparison of DEXA, calipers, BIA and photo estimates, and the visual guide to typical ranges.
Frequently asked questions
Can ChatGPT estimate body fat from a photo?
It can produce an estimate, but we did not find a peer-reviewed study testing general chatbots on this task, so its accuracy is unproven. A general model is not calibrated against measured body fat, and it may not have your height and weight unless you provide them.
Why might a chatbot give different answers for the same photo?
Generative models can word and weigh things differently on each try, and the result depends on how you phrase the question. A structured tool asks the same questions every time, which makes answers easier to compare.
Is it safe to upload a body photo to an AI service?
Check the service's retention and training settings first, because a body photo is sensitive. On ScanBodyFat the photo is used only to produce your estimate, and the privacy policy explains how it is processed and how long anything is kept.
Is ChatGPT useful for anything related to body fat?
Yes. It can explain what body fat percentages mean, help you compare methods, or help you plan questions for a doctor. The limits are in measuring a number from a picture.
Want a number for your own photo?
Free, no signup. Upload one photo and a few details, get an estimated range.
Get your free estimateSources
- Evaluation of three large language models for nutritional content estimation from food images, PMC
- Majmudar et al., Smartphone camera based assessment of adiposity: a validation study, npj Digital Medicine, 2022
- ScanBodyFat privacy policy
More guides
- How accurate is AI body fat from a photo?
- What does a body fat percentage look like?
- Photo vs DEXA, calipers and BIA: which body fat method is best?
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.