Body composition is the metric that actually reflects the thing most people care about when they train: not what they weigh, but what that weight is made of. The scale can't tell the difference between two kilos of muscle and two kilos of fat. A DEXA scan can, but it costs money, requires an appointment, and captures a single point in time.

The interesting shift is that a vision-capable AI model can now look at a well-lit progress photo and produce a body-fat estimate that lands within a few percentage points of a scan — then go further and tell you what to do about it. The photo carries the composition signal; your connected Apple Health data supplies the context that turns an estimate into a plan.

This is exactly what Metrya's Body Vision does. It's worth understanding what's happening under the hood, because the difference between a useful body-composition tool and a flattering toy comes down to a few design decisions — and privacy is one of them.

"The scale tells you a number. A photo, read against your own biometrics, tells you a composition — and composition is the thing you're actually trying to change."

Why a Photo Beats the Scale for Composition

Weight is a single scalar that collapses fat, muscle, water, and glycogen into one number. Two people at the same height and weight can look completely different, and the same person can drop three kilos of water in a week without changing their body composition at all. That's why the scale is such a poor progress signal for anyone doing recomposition — losing fat and gaining muscle simultaneously can leave weight flat for months while the body transforms.

A photo captures what weight can't: the distribution and definition that reveal how much of your mass is lean tissue. Visual body-fat estimation isn't new — experienced coaches have eyeballed it for decades. What's new is that a vision model can do it consistently, at any hour, without an appointment, and cross-reference the estimate against scale data, lean mass from a smart scale, and your training history in the same pass.

Weak signal
Scale weight alone
A single number that can't distinguish fat from muscle from water. Flat during successful recomposition, noisy day to day from hydration and food timing.
Example: weight unchanged for 8 weeks — reads as "no progress," while body fat quietly dropped from 22% to 18%.
Rich signal
Photo + biometrics
A visual composition estimate anchored by scale weight, lean mass, sleep, and training frequency. Captures the change the scale hides.
Example: same 8 weeks read as "visible waist and definition change, ~4% body-fat drop, muscle preserved" — the real story.

What the Model Actually Reads

A body-composition analysis is only as good as the inputs it can draw on. A photo on its own gives a vision estimate; the connected biometric data is what calibrates it and grounds the resulting plan in your reality rather than a generic template. Body Vision uses whatever of these is available:

📷
The photo(s)
Your current physique, and optionally a goal photo. The vision model reads definition, distribution, and the muscle-mass gap between where you are and where you want to be.
⚖️
Weight & lean mass
From a connected smart scale via Apple Health. Anchors the visual estimate to real mass and separates fat from lean tissue.
🏋️
Training frequency
Days per week and environment (full gym, home dumbbells, bodyweight, outdoors). Every recommended exercise has to fit what you actually have.
😴
Sleep
Average hours per night. Recovery capacity sets a realistic ceiling on training volume and muscle-gain rate — under-slept plans fail.
🍽️
Calorie intake
Your daily target, if logged. Grounds the nutrition side of the plan in your actual eating rather than an abstract deficit.
🧬
Biological age
Your bio age versus chronological age, if computed. A younger biological age often signals more recovery headroom for aggressive goals.

The more of this the model sees, the tighter the estimate and the more specific the plan. But none of it is required beyond the photo — Body Vision uses what's available and says so, rather than inventing data it doesn't have.

The Part Most Apps Get Wrong: The Timeline

The easiest way to build a body-composition app that people love and then abandon is to promise fast results. "Reach your dream body in 8 weeks!" sells downloads and destroys trust, because human physiology has hard rate limits that no plan can beat. A tool that respects those limits is less exciting and far more useful.

Body Vision holds itself to evidence-based ceilings when it estimates how long a goal will take:

Change Realistic sustainable rate
Fat loss 0.5–1% of body weight per week, natural and non-drug-assisted. Faster than this and you're shedding muscle and water, not fat.
Muscle gain ~0.25–0.5 kg per month for trained individuals, and slower once you're past the first six months. Beginners gain faster; it doesn't last.
Fat loss + muscle gain together A minimum of roughly 12 months when both changes are significant. The two goals partly fight each other.
Elite / competition physique 1.5–3 years from a typical recreational starting point. A single-digit-body-fat, high-muscle physique is a multi-year project, not a summer.
The honest-timeline design: If you upload a goal photo of an elite athlete, Body Vision explicitly accounts for the muscle-mass gap rather than pretending fat loss alone will close it — and it won't quote a timeline shorter than 18 months for that kind of transformation. The estimate carries a stated ±3% margin. A plan you can trust is one that occasionally tells you your goal is further away than you hoped.

From Estimate to Plan

An estimate on its own is a curiosity. The value is in what comes next: a structured plan across the three levers that actually move body composition — training, nutrition, and recovery. Body Vision returns each as a headline plus concrete items, along with the factors that most influence your specific result.

1
Composition read
Your estimated current body fat, the target for your goal, and a plain-language narrative about the gap between them — grounded in the photo and your scale data.
2
Leverage factors
The two to four things that matter most for you, each tagged as your biggest opportunity, a tailwind working in your favour, or something to watch.
3
Training that fits your kit
Exercises constrained strictly to your environment. A bodyweight-only plan never recommends a barbell; a home-dumbbell plan never sends you to a cable machine.
4
Nutrition & recovery
A nutrition approach anchored to your calorie target and a recovery plan built around your real sleep — because muscle is built in recovery, not just in the gym.
5
Trajectory curve
A projected body-fat path from today to your goal, so the timeline is a curve you can track against, not a single distant date.
6
Re-run and compare
Shoot a new photo weeks later and re-analyse. The visual change plus your updated biometrics show whether the plan is working — the scale can't do that.

The Privacy Question a Photo Raises

A body photo is about as sensitive as personal data gets. Any tool that asks for one has to answer a simple question: where does the image go, and who can see it? For most "AI body scan" apps, the honest answer is that your photo is uploaded to the company's servers, processed there, and stored under a privacy policy you didn't read.

Metrya's model is different, and it's the same model as the rest of the app. Body Vision runs on Bring Your Own Key: you connect your own Anthropic, OpenAI, or Google API key, and your photo goes directly from your iPhone to the AI provider you chose — never to a Metrya server. There is no Metrya backend in the path that could store, index, or train on your image.

Why BYOK matters here specifically: With your own key, the request is between you and the model provider under their API terms — which, for the major providers, explicitly exclude API inputs from training data. Metrya never becomes a data processor for your photo because your health data and your images never touch our infrastructure. More on why BYOK is the right model for health AI →

One practical consequence: Body Vision needs a vision-capable model. Not every model can read an image — the analysis works with providers and models that support vision input (for example, current Claude, GPT, and Gemini vision models), and the app checks capability before it sends anything.

What a Photo Estimate Can't Do

Setting expectations honestly is part of building something trustworthy. Photo-based body-composition analysis has real limits:

Progress You Can Actually See

The reason a photo works as a progress metric is the same reason the scale fails: recomposition is a visual change before it's a numeric one. Fat leaves and muscle arrives in a way that transforms how you look while the scale barely moves. Capturing that visually — and letting an AI quantify the estimate and adjust the plan — turns "I think I'm making progress" into something you can point at.

Used every few weeks, alongside the rest of your Apple Health data, Body Vision becomes a feedback loop: shoot, analyse, follow the plan, re-shoot. The estimate tightens as the model sees more of your data, the plan adapts to what's actually changing, and the whole thing stays on your device and your own API key.

See your composition, not just your weight.

Body Vision estimates your body fat from a photo, maps a realistic timeline, and builds a plan that fits your gear — using your own AI key, so your photo never touches a Metrya server.

Download Metrya Free