AutoRig FAQ: rig, animate and export any 3D model

One upload zone on the home page takes a 3D model (GLB, FBX, OBJ or a ZIP with textures), a picture, a picture of a car, or a text description. Every task gets its own page: one task, one viewer, with the agent chat.

The fast rig builds the skeleton and skin from the geometry in seconds, and the animation clips are retargeted at once. The target is a rigged, animated model within one minute from pickup. The quality checks run in the background of the viewer.

No. The viewer offers to confirm the skeleton in the blueprint, but nothing waits for it: the animation is made at once. Confirm or edit the bones only if you want to.

Every task opens in its own Unity WebGL viewer. It starts on a black scene at once and plays the stages (voxels, erosion, bones, rig) while the mesh loads, then loads the chosen environment scene. When the rig is ready, the animations play from the bar at the bottom, and the AI agent chat sits in the corner.

The right mouse button pans and the middle button orbits; left-drag also orbits, and Shift + left pans. The wheel changes the lens, Ctrl + wheel moves closer. F frames the model, F11 goes full screen, F12 saves a screenshot and F1 lists every key. On a phone, drag with one finger and pinch with two.

Yes. Camera presets frame the face, elbows, hands, knees, feet and heels, so you can check the joints up close. Every control shows its hotkey as a badge, and F1 lists all keys.

Left: model, rig, bones, blueprint, voxels, framing, auto camera and the outliner. Right: the look of the scene. A click switches an icon to its next preset, the last step turns it off, and holding the icon opens its detailed settings. Bottom left: the channels, keys 1–0. A right click opens a menu with this FAQ.

The viewer loads only the scene it needs, after the rig presentation: for example a cathedral, Sponza, a river glade or jungle ruins. Scenes come with baked lighting, reflection and light probes, and colliders.

The agent lives in your task's viewer. It looks at the model, decides what to do next and uses tools: it checks joints, feet and skinning, repairs frames, adds animations, changes materials, light and scene, and renders images and videos of the result.

Type in the chat under the video window in the bottom-right corner, in any language. Ask in plain words: “check the feet”, “make her wave”, “make the eyes glow”. Quick-reply chips suggest next steps, and renders and videos appear right in the chat.

It knows which layer and clip are on screen and whether the bones are shown, and it can drive the viewer for you: play a clip, show a layer, move the camera, change materials, weather and time of day.

No. Your upload stays as it is. Edits are saved as new versions on top of it, and material changes can be undone.

Yes. Each task has one agent chat shared by everyone who opens it, not a separate session per visitor. The agent speaks the language of the person who created the task.

Biped (humanoid), quadruped, equine, bird, serpent, monster, multilimb, hands, primate and vehicle. Each category has its own skeleton template, fit and skin rules, clip library and quality checks.

AutoRig picks it automatically from the shape of the model and what a vision check sees, so you do not choose it. If a model lands in the wrong category, report it with the defect button in the Telegram notice.

Quadrupeds (dogs, cats, bears, cows, pigs) get a quadruped skeleton. Equines (horses, deer, zebras) get a richer one with scapula, elbow, mid-neck and leg profile joints, and planted-hoof gaits: walk, trot and gallop.

Wings are rigged as wings, never as arms. A bird gets its own skeleton and its own clip library.

Yes. The hands category rigs a single hand, gloves or first-person arms with per-finger joints and has hand clips of its own.

Any pose works. When a biped stands on one leg or steps, the leg on the floor sets the bone lengths for the other, so both legs get the same lengths. If the file already has a skeleton or an animation, its own rest pose is used first.

Yes. Drop a picture of a car on the home page. The vehicle conveyor segments it into body, wheels, doors, hood, trunk, hatches, glass, lights and bumpers, models each part as its own 3D object and assembles a rigged car.

Every part is isolated and restored as a whole object, and the viewer lists them in the Parts panel. For each part you accept it, redraw it or order it as a 3D model. The body is drawn as a bare shell without doors, hood and glass, so those parts open and move on their own.

Each wheel is normalised to a cylinder collider that matches its own shape. The body is corrected to left–right symmetry with a free-form deformation (FFD) lattice, and the doors are hinged.

Yes. The cabin has a seat, a steering wheel, pedals and a gear stick, and a rigged driver is seated at them by IK.

The free farm is the default. Paid 3D clouds (Tripo and Hunyuan cloud) run only when you press their explicit button.

The fast rig builds the skeleton and skin from the model's geometry in seconds, using the template of its category, and the animation clips are retargeted at once. Quality checks (joint placement, skin tears, limb collisions, numeric deformation QA) run in the background of the viewer. Humanoids get a standard humanoid skeleton that animation tools and game engines understand.

Press F8 or the bones icon to see the skeleton over the model. Play a clip, press Space to stop on a frame and step with the arrow keys. Channel 0 colours the body parts, and the outliner (9) hides or isolates any part.

Yes. Press B for Blueprint mode and move the bones; symmetry mirrors your edit to the other side. The rig is rebuilt 15 seconds after your last edit, or at once with the re-rig button, and the correction goes into every animation. You can also ask the agent to check the feet, the knees or a single frame.

Every model that once broke a rig becomes a regression test, and every engine release has to pass these autotests before it goes live. Rig quality comes before speed.

People, stylised and anime characters, animals, birds, creatures, hands and cars. Upload GLB, FBX, OBJ or a ZIP with textures, or generate a model from a picture or text.

A rigged character comes with ready motions such as idle, walk, run, jump, wave, dance, punch and look around; every other category has its own clip library, for example walk, trot and gallop for a horse. They play from the bar at the bottom of the viewer, and the agent can add more on request.

For each category, reference videos of the gaits and actions are generated and the motion is extracted once. The clips are then retargeted automatically onto every new rig, with a numeric QA check on each clip. Procedural clips are the fallback.

Retargeting moves a motion made for one skeleton onto yours while keeping your character's proportions. Your joint corrections are carried into every retargeted clip.

Yes. Press Space to stop the clip, find the frame on the slider and press the wrench: the frame is smoothed with its neighbours and flips in the clip are repaired.

Yes. Custom clips can be added in the viewer to the task's clip set, and they are retargeted onto your rig like the built-in ones.

Yes. Pick any object, material or body part in the outliner (9) or click it in the scene, then change its colour, metalness, smoothness and glow. The eye icon hides parts so you can look inside.

Describe the change and the agent makes it: recolour clothes, make eyes or armour glow, try other materials. Every material change can be undone.

Yes: the post-processing look, depth of field, volumetric light, weather, time of day, stages such as a studio or a cathedral, and a radial pulse on Alt. Each is one click on the right toolbar.

Yes. Upload a picture: AutoRig builds a 3D model from it, rigs it and opens it in the viewer with animations. A picture of a car goes through the vehicle conveyor instead.

Yes. Write what you want to see; the character is generated from the description, then rigged and animated the same way.

Yes. Send a video and the agent uses it as a reference for the character and its movement.

Yes: renders, turntables and short videos of the character in a scene. They appear in the chat next to the viewer.

Use the keys 1–0 or the icons in the bottom-left corner: 1 full shading, 2 albedo, 3 normals, 4 metallic, 5 roughness, 6 emission, 7 object IDs, 8 material IDs, 9 the outliner, 0 body parts.

Each separate object or material gets its own flat colour, so you see at once how the model is split and which parts share a material.

Yes. A model's own textures are never lost: materials, textures and UVs are kept through rigging, and the PBR maps (base colour, normal, metal-roughness and emission) show up in the channels.

It gets PBR textures automatically (albedo, roughness, metallic, normal and emissive), made from the model's own projections. The textured model is what the vision checks and the previews use.

The rigged, animated model downloads as GLB and FBX. USDZ, a ZIP bundle, a Unity package and a Blender file are offered too, and static copies are available as OBJ.

Yes. The FBX imports as a humanoid in Unity and as a skeletal mesh in Unreal, and the GLB opens in Blender, three.js and other web viewers. A ready Unity package and a Blender file come from the same task.

Yes, if you own the rights to the source model. AutoRig adds the skeleton, skinning and animations and claims no ownership of your model.

Yes. Tasks can be created, tracked and downloaded through the API; the developer page has the details.

The site is available in English, Russian, Chinese, Hindi and Persian (right to left). English pages have no prefix; the others live under /ru/, /zh/, /hi/ and /fa/.

The Telegram notice of a finished task has defect buttons, for example a wrong model transform, a torn mesh after skinning or elbows that are too wide, and you can reply with text too. A report is recorded and checked, and the fix goes into the engine for the whole category, not into one model.