AI back analysis: how one photo becomes six zone scores and a plan

AI back analysis is a vision model reading one photo of your back and returning numbers instead of an opinion. In BackMaxx it runs like this: selfie camera, five-second self-timer, one shot with your back to the lens. What comes back is a Back Score from 0 to 100, a score for each of six zones, a score for each of eight muscles, the zone you are furthest ahead in, the two you are furthest behind in, and a training week rebuilt around the lowest of those numbers.

The photo is analyzed transiently and never stored on a server. Nothing about your body composition is estimated, because a photo cannot supply it.

What AI back analysis actually is

Two words are doing separate jobs here. The rating is the output — one number, one rank, one place on a scale. The analysis is the decomposition: which parts of the back produced that number, and in what order they deserve attention. The rating side of the same scan is the headline. This page is the machinery under it.

What gets analyzed is visible muscle development seen from behind. Shape, fill, separation, the outline of the torso, left compared against right. That is the entire domain, and it is a deliberately small one.

What does not get analyzed: body fat, weight, girth, strength, or any measurement of the body. An analysis of a photo produces an estimate of how developed muscle looks, and BackMaxx makes no claim past that.

How the analysis runs, tap by tap

The loop is one photo long, and most of the work happens before the shutter fires.

1. Set the shot up first. Prop the phone at chest height with the screen facing you, pick even light and a plain background, frame yourself, then turn around with your arms relaxed and your full back inside the frame. Shirtless or a fitted tank shows the most.
2. Tap, turn, wait five seconds. The self-timer is the whole trick. The selfie camera lets you frame yourself while you can still see the screen, and the spoken countdown lets you hear the shutter with your back already turned. One shot, not a burst.
3. Or skip the camera entirely. Any back photo already sitting in your library can be imported and run through the same analysis.
4. The photo travels alone. One JPEG goes up, plus three answers you gave once during setup: your goals, your experience level, and whether you train at home or in a gym. No name, no account, no email address, no device history.
5. What comes back is structured, not written. A Back Score, six zone scores, eight muscle scores, your strongest zone, your first and second priority zones, and a single insight line that reads your stated goals back against the numbers.
6. A photo the model cannot read is refused, not guessed at. A frame that is too dark, too tightly cropped or not a back returns a reason instead of a score invented to fill the gap.
7. The plan is assembled on the phone. Once the numbers land, the training week is generated locally from them and keeps working with the phone in airplane mode.

What the model reads in each of the six zones

Six zone scores come out of one frame because each zone is a different feature of the same silhouette.

Each read maps onto what that zone buys you in the app’s own words. Lat Width is “the V-taper, width from the armpit down”. Upper Back is “thickness and the 3D look from the side”. Rear Delts are “rounded shoulders from behind — the detail that pops”. Lower Back is “the Christmas tree, and the base every heavy pull needs”.

Symmetry is the odd one out, and not only because no exercise in the corpus claims it as a primary zone. Every other score answers how much is there. Symmetry answers whether the two halves agree — “left equals right, balance reads as quality” — which is why it can come back low on a back whose other five zones are all fine.

Which muscles sit under each zone score

Under the six zone scores sits a second layer: eight muscle scores, each belonging to exactly one zone. The muscle map is where a zone number stops being a region and becomes a target.

That last column is the reason two low muscle scores are not equally easy to act on. A low Lats score has 38 of the 77 movements to draw from, at every equipment level from a doorway to a cable stack. A low Infraspinatus score has 8, and they are the same short list of face pulls and rotation work every time — a muscle that “trains as a rotator, not a puller”, at light load only.

The gap between zone and muscle detail also matters for how you read a scan. A Lat Width score of 51 does not say whether the Lats or the Teres Major is holding it down, and those two want different work: full-range vertical pulling for one, wide-grip and straight-arm work that drives the elbow down and back for the other.

From a low number to a different week

A score is only worth taking if something changes because of it. Here is exactly what changes.

The three lowest zone scores become the week’s priorities, worst first, and the session is built from that ordering.

Across a week the lead rotates. The first session opens on your worst zone, the second on your second worst, the third on your third. You choose 1, 2 or 3 sessions a week at 15, 30, 45 or 60 minutes, and the time budget decides how much fits: three lifts at 15 minutes, four at 30, five or six with a warm-up and a cool-down at 45. The budget also caps how advanced a movement can be, and nothing repeats across the week.

Take Rear Delts coming back lowest. That zone then opens every session it leads, and the engine has 8 movements carrying Rear Delts as their primary zone to fill two slots from — Cable Rear Delt Fly, Reverse Pec Deck, Rear Delt Fly and Bent-Over Rear Delt Raise in a gym, Band Pull-Apart at 3 sets of 15 to 20 and the banded fly at home. The prescription matches the muscle: small, high frequency, light weight, strict form, more sets than you think.

Symmetry behaves differently again. Nothing in the corpus is a Symmetry exercise, so a low Symmetry score pulls the 8 movements that carry it as a secondary zone, and every one of them is one-side-at-a-time work: One-Arm Dumbbell Row, Single-Arm Cable Row, Meadows Row, Kettlebell Single-Arm Row, Commando Pull-up, Single-Leg Romanian Deadlift, Landmine Romanian Deadlift and Bird Dog. A zone that stays far below the rest eventually earns a block of weeks rather than a slot.

That is what the five focus programs are for. V-Taper Sprint spends 4 weeks leading with Lat Width, Thickness Block spends 4 on Upper Back, Traps & Rear Delts takes 3, and Pull-Up Progression takes 6 built around a bar. A running program overrides the lead zone while it lasts; when its weeks are up it drops away on its own and the scan takes the lead back.

Why analysis beats a mirror

A mirror does not show you your back. That is not a figure of speech — the one area you train blind is the one that decides how you read from behind, and the bathroom mirror shows you the front of someone who has been studying his own front for years.

A phone photo solves the seeing problem and leaves the comparing problem untouched. Two photos six weeks apart, shot under different bulbs at different distances, get compared by memory and mood.

The analysis fixes the scale. The same six numbers come back every time, read by the same three bands: 80 and above is strong, 60 to 79 is developing, below 60 is lagging. A zone that moves from 54 to 63 has crossed a band, and no argument about the lighting changes that arithmetic. The nine rank bands, from Newbie under 38 to GOAT at 91 and up, do the same job for the total.

Ranking is the other thing eyes cannot do. Ordering your own six zones worst-first and honestly is harder than scoring any one of them, because everyone over-rates the zone they enjoy training and forgives the one they skip.

What one photo of a back cannot tell you

The frame simply does not contain some things, and the analysis should not pretend otherwise.

• Load. Nothing in a photo reports what you row, pull or hinge. Visible development and strength travel together, but the model reads shape.
• Body composition. No body fat figure, no weight, no circumference, no measurement of any kind. BackMaxx does not estimate them.
• What sits under a soft layer. Separation is read from shadow, and shadow needs both light and definition. At higher body fat the detail scores read conservatively, because the detail genuinely is not visible.
• Movement smaller than the noise. Backlight flattens the shadow that separation is read from, a flared lat spread inflates Lat Width, and standing a step closer to the lens changes every proportion in the frame. Two or three points between scans is usually camera, not muscle.
• Anything about the front, and anything clinical. Six zones, eight muscles, one view, one purpose: training.

Those limits are why the sane cadence is scanning weekly and reading monthly. Same spot, same light, same distance, arms in the same position, and the noise mostly cancels out across four or five scans while real change does not.

What leaves your phone and what stays on it

Here is the complete inventory, because on a page about photographing your own back it is the first thing worth stating.

Deleting a back photo on your phone therefore leaves no copy anywhere else. There is no account to close, no email address on file, no ad network in the app, and no sign-up standing between installing it and your first scan. The end-to-end pipeline spells out where each piece of the scan is stored.

AI back analysis vs a whole-body AI body scan

The two categories look similar in an app listing and answer different questions.

The trade is deliberate. A whole-body number averages chest, arms, legs and back into one figure, and averaging is precisely the operation that hides a lagging back — the area you never see is the area a single score is happy to bury. Decomposition is the opposite move.

Narrowing the subject also narrows the claim, which is the honest part. One photo taken from behind contains almost everything there is to know about how developed a back looks, and almost nothing about the rest of a body. Scoring only the thing the photo actually shows is what makes the number worth taking again next week.

Find out where your back actually stands

One photo, a Back Score out of 100, six zones and eight muscles ranked — then a plan built on whatever came out lowest.

Download on theApp Store Download on the App Store

Free to download. iPhone, iOS 26 or later.