Bodies in three dimensions: from plausible to measured

For twenty years the digital human body was mostly an animation problem: a mesh that could be posed, skinned and rendered convincingly. The same body is now asked to do much more. It decides whether a garment fits before it is sewn, which size a shopper is offered, how a person moves through a crowd model and what a health measurement means. Those jobs share one requirement that animation never had. The body has to match a real one, in centimetres, and say how closely.

2025 moved the field a long way on breadth and openness. The step that matters next is measurement.

Open body models grew up

Parametric body models describe a body with a small set of numbers, shape and pose, that drive a template mesh. Widely used research models were fitted to collections of 3D scans. Two 2025 releases widened what is available.

In November 2025 NAVER LABS Europe released Anny, a parametric model that covers the full human lifespan, from infants to elders, in a single model under the Apache 2.0 licence, with no registration or gated download. Anny is built without 3D scans. It draws on anthropometric knowledge from the MakeHuman community and is calibrated against WHO population statistics, with interpretable controls for age, height, weight and muscle on a base mesh of more than 13,000 vertices, 163 bones and 564 blendshapes. Its authors also produced Anny-One, a synthetic set of around 800,000 images, and report that human mesh recovery models trained on Anny can match those trained with scan-based body models.

The same month Meta published SAM 3D Body, which recovers a full-body mesh, including hands and feet, from a single image, and can be guided with keypoints and masks. It is the first model to use the Momentum Human Rig, a body representation that decouples skeletal structure from surface shape. Meta states that both are open source. Separating the skeleton from the surface is a practical gain: bone lengths and joint positions can be set from measurement while soft tissue is fitted on top.

Together these releases make a capable body representation free to use for anyone, at any age, and make recovering a plausible body from a photograph a routine research tool.

Plausible and measured are different targets

A body recovered from a photograph is plausible: it agrees with the image. Whether its waist matches the person's waist to a centimetre is a separate question, and the answer depends on the camera, the clothing, the pose and the model's priors. Calibrating against population statistics makes a model representative of people in general. It says nothing about a particular person.

Consumer body scanning shows the size of the gap. A 2025 study compared a free smartphone 3D scanning app with flexible tape measurement in 32 adults and found significant differences between app and tape at every site. Absolute error was smallest at the hip, 6.3%, and calf, 6.8%, and largest at the forearm, 14.8%, and abdomen, 13.7%. The authors advise caution where accurate measures at particular sites are needed. A 6% error on a 100 cm hip is six centimetres, about the whole of the ease a fitted garment may allow at that point.

Errors like this are fine for some uses and disqualifying for others. They are only a problem when nobody says they are there.

Why centimetres matter commercially

Sizing remains the most common reason clothes come back. Akeneo's 2025 Consumer Returns Report found that sizing was the most frequently cited reason for returns, raised by 58% of consumers. Every one of those returns is a garment made, shipped and handled twice, and since July 2026 large companies in the EU may no longer destroy unsold clothes.

The fix is a chain that holds its accuracy end to end: a body fitted to real measurements with its error stated, a pattern drafted for that body, a cloth simulated with measured properties, and a fit read off the result as numbers, ease at bust, waist and hip, stretch along each seam, rather than judged from a picture. Each link can be checked. The size recommendation at the end is only as good as the weakest of them.

What exact body representation enables

Once a body is held to its measurements, several things become possible that a plausible body cannot support.

Fit before sampling. A garment can be judged for a specific body, or a whole size range of measured bodies, before cloth is cut.

Size ranges built from people. Grading rules can be tested against bodies that span real populations, including children and older people, now that open models cover the whole lifespan.

Health measurement with context. A body measurement taken over time is useful only if the change is larger than the measurement error. Stating that error with every reading is what turns a number into evidence.

Digital humans that are someone. A lifelike avatar draped in its own measurements shows how a garment falls on that person, rather than on a generic figure.

Where Livgens fits

Livgens represents the thing exactly, in sound, light, matter and motion, and Living Worlds runs it forward through its lifecycle. Bodies are among the things Livgens represents: a person's own measurements, and bodies fitted to any size, across Health and Render. In Render's fashion house case study, a body was fitted to four measurements, height, bust, waist and hip, to within a tenth of a centimetre. A parametric pattern was drafted for that body with the ease the design asks for, draped on the GPU with NVIDIA Warp, and the fit read off as ease at bust, waist and hip and stretch along every seam, so a tight bust or a pulling seam shows as a number. The arms are in the rest pose, not a styled one.

The case study is plain about what comes next. Digital humans, lifelike people draped with their own bodies and measurements and shown live with their own skin and hair, are in progress; garments that move with the body, and garments settled cleanly on lifelike people as they are on the standard bodies today, are listed as next. Living Worlds runs the lives around a product forward through their lifecycle, the households that buy and how their wants change, and the fit read from the drape is a measured input those simulations can use, held as unvalidated until it is scored. Render is in early access, and Health is live.