Animal library · first model in production

Every animal, rebuilt from the evidence.

Each species starts from a real 3D scan or published measurements, a skeleton measured to the animal and photographs to check against. The first one through is the red fox.

Red fox, studio turntable, November 2026. Derived from "test scan of taxidermy fox" by Jakob.Normand, CC BY 4.0, changed by us: rig, skinning, fur shells, eyes, colour de-lighting, levels of detail and animations.

The drawer

Species in the pipeline.

One animal at a time, each from its own real scan. Recolouring one species into another does not work, so every species waits for a scan of its own.

The red fox model on studio grey.
No. 001Mammal
Vulpes vulpes

Red fox

In production

From a CC BY scan: rigged, furred, with 105 movement clips generated on its measured skeleton. The paws and some leg poses are being fixed before it can sell.

The common raven model in Unreal Engine on studio grey.
No. 002Bird
Corvus corax

Common raven

Feathers passed

From a CC0 museum scan. The first bird: its feather layer beat the bare scan, and a species-recognition model places it with the ravens in all six test views, in Blender and in Unreal. Eye size is measured from real photographs.

The moose bull model in Unreal Engine on studio grey.
No. 003Mammal
Alces alces

Moose (Eurasian elk)

In testing

From a CC BY scan of a bull, matching in Blender and in Unreal. Walk and trot are in testing; its gait numbers are still estimates until measured moose data is found.

HeldMammal
Canis lupus

Grey wolf

Held

Built from measurements and photographs over 18 versions, before the scan route proved faster. Held until the fox is done.

What one scan does

One scan beat everything we had built by rule.

The same red fox, scored by a species-recognition model (BioCLIP 2) choosing among 18 look-alike species. Left: the fox we built from rules and trait data in October. Right: one raw scan of a mounted fox, before any of our work on it.

Our rule-built fox from the side: bright orange, fluffy, cartoon-like.
Built by rule, Oct 2026 · side view
The raw scan of a mounted red fox from the side: natural russet coat, black legs, white chest.
One raw scan · side view
Built by rule
5.6%
One raw scan
31.2%
Scan with its fur
74.9%
Real photographs
75.1%

Chance a species-recognition model (BioCLIP 2) names it a red fox among 18 look-alikes, over six standard studio views. The furred fox is the mean of five lighting runs (±1.7). The rule-built fox was lit in a different studio, so read the direction more than the exact gap. Real photographs (156 of them, in their own scenes) are shown for reference: they are the bar. Passing a species check is not the same as looking right, so it is one test among several. The face, where taxidermy sculpts the muzzle, is where partner data matters most.

What one scan became

From a free test scan to a fox that trots.

The red fox started as one scan of a mounted fox, shared under CC BY 4.0. Everything after that is our pipeline, and the same pipeline runs for the next species.

01The raw scan of a mounted red fox, front three-quarter, on grey.

The scan

The mount's shape and its real surface colour, as photographed. Stiff, with no skeleton inside.

02The fox model with a full coat, three-quarter view, on grey.

Skeleton and coat

Bones measured inside the scan, a rig, and fur grown from the scan's own surface. In a studio test the coat lifted the species score from 0.51 to 0.78.

03The fox mid-trot on a forest floor, spruce trunks behind.

Motion

105 clips, from walk and gallop to sitting, growling and yawning, generated on the fox's own skeleton at real speeds and strides.

Still open before it can sell: the paws, and legs that push through the coat in some poses. Both are on the build progress page.

How a model is made

One pipeline, run again for every species

The steps are code, so each species goes through the same checks, and a fix to the pipeline improves every animal built on it.

  1. Scan

    A 3D scan of the real animal, a mount or a skeleton, with its licence recorded. Where no scan exists, published measurements set the shape.

  2. Skeleton

    Bones measured inside the scan, with the same bone names across every species of a body plan.

  3. Skin and rig

    Weights solved from poses, so the body bends at the joints without folding or poking through.

  4. Coat

    Fur grown from the scan's own surface colour, kept only if it scores better than the bare scan.

  5. Gaits

    Walk, trot, lope and gallop generated on that skeleton at the species' real speeds and strides, plus behaviours such as sitting, growling and yawning.

  6. Checks

    Every clip is checked for feet sliding, body poking through the coat, and how strongly a species-recognition model names the render.

  7. Export

    The Blender source file, levels of detail and an Unreal-ready export.

The source card

Every number says where it came from

Each animal carries a record of its measurements, the photographs it was fitted to and the licence of each one. When nobody has measured something, the card says so instead of guessing quietly.

This is the card for the grey wolf, which we built from measurements before the scan route proved faster. Its figures come from the wolf account in Heptner and Naumov.

Canis lupusHeld behind the fox
Grey wolf · western Scandinavia · about 11,700 years ago
Body mass, adult male
40–45 kg
Head and body
118–137 cm
Tail
42–52 cm
Hind foot
25–26 cm
Ear
12.7 cm
Skull, greatest length
257–267 mm
Model skull length
262 mm
Reference photographs
22, licence each

Heptner V.G. & Naumov N.P. (eds) 1998, Mammals of the Soviet Union, vol. II part 1a, wolf account. No Ice Age Scandinavian wolf has been measured, so these are living Eurasian wolves, marked on the card as an analogue. The model's skull sits between the tundra and Taimyr male means. Spine proportions were measured off mounted skeletons.

Who it is for

Anyone who needs the animal to be right

Game studios

Hunting, survival and open-world games where players know what a reindeer looks like at a run.

Film and VFX

Background wildlife that holds up in a documentary or a period drama.

Museums and education

Animals of past ages, including extinct ones, built to the measurements.

Research and AI training

Synthetic pictures of named species for training wildlife detectors on camera traps and drones.

Planned

What each model will include

Blender source

The full working file, editable and engine-free.

Standard skeleton

The same bone names across a body plan, so animations and tools carry over between species.

Gaits

Real-speed movement cycles for the species.

Levels of detail

From close-up hero to far herd.

Source card

Every measurement and photograph, cited.

A clear licence

Built only from open sources and our own work, so the rights are clean.

Not on sale yet, and pricing is not set. The build progress tab shows where the first models are.

Why we need help

Open data runs out exactly where realism starts.

Trait tables give us the size and shape of every species. Looking real takes a captured surface: a scan of the actual animal. Scroll through what exists.

About 47,700 animal species have a rough draft in our tables. Each dot is one of them.

About 155 open scans. That is how many whole-body, textured scans of vertebrates we found that we are allowed to build a sold model from. Give each one a dot and that is how far they reach: about one species in 300.

About 15 of those are standing mammals like our fox. Most of the rest are birds.

No wolf. We searched every open collection we could reach. Without a scan we built 18 versions of a grey wolf from photographs and published measurements, and none looked right. The wolf is on hold.

Buying doesn't fix it. None of the paid stock licences we checked lets anyone resell a changed model. The missing scans, footage and measurements sit with the people who care for and study these animals.

Counted on Sketchfab in November 2026 by looking at every candidate, the biggest open pool we searched. MorphoSource had no open, sellable vertebrate meshes. Lists and method are in the build records.

What we are looking for

The data open sources don't have.

Every animal we build is checked against the real one. These are the places where our checks fail for lack of data, and where a partner's collection, footage or notes would make the biggest difference.

Close view of the red fox model's head and chest: russet face, dark eye patches, white throat.
Faces are the hardest part. This head comes straight from the scan; the muzzle still reads as a mounted fox's.
Scans

Mounts, skulls and skeletons

One phone photo-scan of a mounted fox gave us its shape and colour in a single step. Scans of other species would do the same for them.

Video

Animals moving

Side-on walks, trots and gallops on known ground. We measure stride, speed and leg angles from the frames and build the gaits from those numbers.

Close-ups

Feet, faces and mouths

Photographs or casts of paws, pads, teeth and lips. No open dataset covers these well.

Records

Measurements and notes

Body masses, limb lengths, coat samples and observations from keepers, vets and field researchers.

Coverage

Rough drafts for about 47,700 species.

A generator turns open trait data into a rough draft of every species: right size, right proportions, right body plan. Drafts are the starting point the pipeline refines. They are not the finished product.

GroupSpecies draftedBuilt from
Land mammals4,431PHYLACINE, COMBINE
Bats1,162PHYLACINE, COMBINE
Sea mammals128PHYLACINE, COMBINE
Kangaroos and kin106PHYLACINE, COMBINE
Birds11,525AVONET, AVOTREX; plumage colour from the HBW plate set (CC0)
Fish11,502FISHMORPH, Rabosky et al.
Reptiles12,060ReptTraits
Amphibians6,756AmphiBIO
Insects, spiders, crabs, shellfish80A chosen list: eaten, used, traded, harmful or seen every day

Work with us

Have scans, footage or measurements of an animal? Need one built?