What Pip can and cannot measure.

Four-angle photo method

Pip asks for front, left, back, and right standing photographs. On iPhone, Apple Vision and device motion help check framing, pose, outline, and phone stillness during capture. The photos are then used for AI body analysis and to fit a 3D exterior body model.

Photographs do not see internal tissue. Clothing, posture, camera height, lighting, hydration, recent meals, and training can materially change appearance. Repeat scans are most useful when those conditions are similar.

AI body analysis

After the user agrees, Pip prepares background-removed, head-masked copies of the four photos and sends them with limited body context to OpenAI through Pip's Railway service. The model returns a bounded set of exterior shape observations, plausible body-fat bands, and regional evidence. Pip validates the response before using it.

The request tells the model not to infer identity, race, ethnicity, diagnosis, health status, pregnancy, medication use, treatment response, strength, performance, or athleticism from the photos. The result may still be wrong and is not a diagnosis or direct measurement.

Fitted 3D body

Pip separately uses the four full photos to create a fitted exterior 3D body through its private Modal processing service. It combines the model output with photo outlines and fit checks, and withholds a fitted result when the evidence does not meet the release's display rules.

A fitted body is a visual representation, not a scan of internal anatomy. It can smooth, miss, or exaggerate body shape and should be compared only with other Pip check-ins captured under similar conditions.

Displayed body-fat and waist estimates

Pip may show an estimated body-fat range and an estimated waist value. Body-fat is presented as a range because the photo evidence does not support a precise percentage. Waist is inferred from the fitted exterior model and entered height; it is not a tape-measure reading.

Pip has not completed an independent criterion study against DXA, a four-component body-composition assessment, or technician tape measures. Individual error, subgroup performance, repeatability, and sensitivity to real change are therefore not established. Do not treat these values as clinical measurements.

Illustrative twelve-week path

Pip may show a broad twelve-week visual path using the current fitted body, user-entered context, and conservative scenario assumptions. If treatment context is present, the illustration may use broad published group-level patterns; it does not predict that person's response, prove treatment effect, or recommend a product or dose.

The path is an illustration of one possible direction, not a promise, forecast, clinical comparison, or expected outcome. Actual change can be smaller, larger, different, or absent.

Ask Pip answer method

Ask Pip first builds a supported answer from the user's saved record on-device. When the user separately agrees to AI answers, Pip may send the question and that short evidence summary to OpenAI to improve the wording. The local answer remains available if the service is unavailable.

Medication, dosing, missed-medication, urgent-symptom, diagnosis, causation, and treatment-advice questions are kept out of generative AI and receive fixed safety guidance. Ask Pip is a record-reflection tool, not a medical-advice service.

What published research supports

Research shows that purpose-built photo or 3D optical models can estimate body composition when trained and validated against a criterion method. Examples include a smartphone-photo validation study and external validation of 3D optical imaging. These studies support the field's feasibility; they do not validate Pip or transfer their accuracy to Pip.

• Majmudar et al.: smartphone camera adiposity validation against DXA
• Harty et al.: machine learning and 3D optical imaging
• Cerqueira et al.: systematic review of Body Adiposity Index validity

Safe interpretation

• Treat every shape and number as an uncertain AI estimate.
• Compare check-ins only when capture conditions are similar.
• Do not infer diagnosis, health status, or treatment effect.
• Use a qualified clinician and appropriate reference method for health decisions.

For data handling details, read the Privacy Policy.