Let's start with the conclusion. A 3D model that AI generates from a single drawing or photo cannot be used as a VRChat avatar as it is. VRChat humanoid avatars have to use Unity's Mecanim humanoid configuration, and the official documentation states that "humanoid avatars must have head, hands, and feet bones mapped" (source: VRChat Creation - Rig Requirements, as of July 2026). AI-generated 3D models are static meshes with no bones inside, so they don't meet this requirement.
The same goes for converting to VRM. VRM is a format designed for humanoid avatars, so simply "converting" a model with no skeleton won't turn it into an avatar. What sits in between is a production stage called rigging and weight painting, and that is modeling work, not file conversion. In this article, we spell out exactly what is missing and how much, and then sort out the uses that are still realistic. If you're just getting started with VRChat, reading Can't Make a VRChat Avatar? A Beginner's Path, Sorted by Where You Get Stuck first will give you the big picture.
*Figmee currently lets you upload your drawings and illustrations to generate and download figurine-style images and 3D model data (GLB / 3MF). Sales of physical 3D-printed items are Coming Soon.
What's Missing
Line up what an AI-generated model has against what a VRChat avatar needs, and the gap becomes obvious.
| Item | AI-generated 3D model (GLB) | What a VRChat avatar needs |
|---|---|---|
| Mesh (shape) | Yes | Yes |
| Textures and materials | Yes | Yes (but shaders need adjusting) |
| Bones (skeleton) | None | Required (including head, hands and feet) |
| Binding between bones and mesh (weights) | None | Required |
| Facial blend shapes | None | Needed if you want lip sync or expressions |
| Avatar setup component | None | VRC Avatar Descriptor is required |
| Polygon count and performance optimization | Not tuned | Performance rank criteria apply |
Every row marked "None" is extra work you'll have to do. The stage where bones are added is called rigging. We explain how it works in What Is Rigging? A Gentle Guide to How 3D Models Move and When You Need It, but getting a single character to the point where it moves without breaking is a skill area of its own within 3D production.
VRChat's Rig Requirements (in Detail)
Here are the requirements listed in the official documentation, ordered by how much they matter once you actually start working.
- Required bone mapping: In addition to head, hands and feet, the Pelvis, Spine, Chest, Neck and Shoulders must be mapped.
- Hierarchy order: The shoulder and neck bones must be direct children of the chest bone. Likewise, LowerArm must be the first child of UpperArm, Hand the first child of LowerArm, LowerLeg the first child of UpperLeg, and Foot the first child of LowerLeg.
- Finger bones: For full IK (inverse kinematics) to work, the thumb, index and middle finger bones need to be mapped. Without them, according to the official documentation, "your avatar will not be able to crouch, and its feet will not automatically step."
- Eye bone axes: "The Y axis must point up and the Z axis must point forward."
- Pose for full-body tracking: The documentation says "the angle between the hip and thigh bones should be close to 180 degrees," measured in a T-pose.
- Toe bones: Not required, but mapping them enables tiptoe movement and makes foot stepping look better.
Seeing it as a list makes it clear: this is not "roughly put some bones in." These are structural requirements. If you add bones to an AI-generated model after the fact, you'll be placing them one by one to match this hierarchy, then adjusting how strongly each vertex of the mesh follows each bone (the weights).
What to Know About VRM and UniVRM
VRM is a 3D file format for humanoid avatars, built as an extension of glTF 2.0. The standard implementation for Unity is UniVRM, released as open source under the MIT license.
Here are the facts stated in the official repository (source: UniVRM (GitHub), as of July 2026).
- Supported Unity versions: From v0.128.0 onward, Unity 2022.3 LTS or later is required. Older Unity versions (2021.3 / 2020.3 / 2019.3 / 2018.4) are covered by past releases but are listed as outside official support.
- Supported formats: Both VRM 1.0 and VRM 0.x, plus glTF 2.0 (
.glb/.gltf/.zip) and VRM-Animation (.vrma). - Installation: You can either download a unitypackage from the GitHub releases or use UPM by adding a git URL in the Package Manager. With UPM, VRM 1.0 requires
com.vrmc.gltfandcom.vrmc.vrm.
In other words, UniVRM can load glTF / GLB. But "it loads" and "it becomes a working VRM avatar" are two different things. VRM is a standard for humanoid avatars, designed on the assumption that the model carries information such as bone structure, gaze and expressions. Loading a mesh with no bones won't generate any of that.
Can You Bring a VRM File Straight into VRChat?
This is an easy point to mix up. The avatar creation and upload process described in VRChat's official documentation goes like this: set up a Unity project and the SDK with the Creator Companion, place the model in Unity, add a VRC Avatar Descriptor, then build and upload from the SDK (source: VRChat Creation - Creating Your First Avatar, as of July 2026). The process is split into eight steps.
- Step 1 - Choose a 3D model
- Step 2 - Set up the VRChat SDK
- Step 3 - Import the model into your project (with the rig configured correctly)
- Step 4 - Place the model in the scene
- Step 5 - Add an Avatar Descriptor
- Step 6 - Check the avatar for problems in the Builder tab
- Step 7 - Build and upload
- Step 8 - Use your avatar
This official flow has no step for uploading a VRM file directly. VRM is a format for platforms that support VRM, and bringing a model into VRChat means going through this Unity + SDK route anyway. The idea that "once it's VRM, it works in VRChat" gets the order backwards.
There's one more important constraint. The Unity version VRChat uses is fixed at 2022.3.22f1, and the official documentation tells you to stay on it, warning that upgrading to a newer version will make your uploaded content fail to load in VRChat (source: VRChat Creation - Current Unity Version, as of July 2026). The safest choice is to let the Creator Companion install Unity for you.
The Full Workflow and Realistic Time Estimates
If you aim to turn an AI-generated model into an avatar, the workflow looks like this. Times vary a lot with your 3D experience, so treat them as rough guides only.
- Clean up the mesh (1–3 hours). Generated models don't have a face layout suited to editing, so you tidy them up with remeshing or retopology. Skip this, and weight painting later is much more likely to break down.
- Adjust the shape for avatar use (a few hours or more). Bring the arms closer to a T-pose, separate the fingers, split hair and accessories into manageable parts, and so on. Models generated from a single drawing often have fused fingers, and shaping them here is a prerequisite for adding finger bones.
- Rigging (adding bones) (a few hours to a few days). Place them so they meet VRChat's hierarchy requirements.
- Weight painting (a few hours to a few days). Repeatedly check things like whether the shoulder collapses when the arm bends. In practice, this is where most of the time goes.
- Create facial blend shapes (optional, a few hours or more). Needed if you want lip sync or changing expressions.
- If you want VRM, export with UniVRM (an hour or more). Install and configure UniVRM in a Unity 2022.3 LTS or later environment.
- Configure for VRChat (1–3 hours). Create a project with the Creator Companion, import the model, set up the Avatar Descriptor, then build and upload.
Steps 3 and 4 take up most of the total. This is not "conversion work"; it's "production." Put the other way around, you might use AI to save on modeling effort, only to spend about as much time on the avatar stage anyway. Understanding this from the start keeps your plans from falling apart halfway through.
Uses That Are Realistically a Good Fit
So does that mean AI-generated 3D models are useless in VR and social media contexts? Not at all. Outside of avatars, there are plenty of uses that work with no extra steps.
| Use | Extra steps | How well it fits |
|---|---|---|
| Rotating it to look at, showing it as a piece of work | None | High |
| Subject for photos and screenshots | None | High |
| Objects and props placed in a world | Almost none | High |
| Backgrounds and display pieces in games or apps | Almost none | High |
| A rough base or starting point for sculpting | None | High |
| A static standing pose of a character | Few | Medium |
| An avatar you move yourself | Many (rigging required) | Low |
Objects placed in a world don't need bones because they don't have to move. This is where AI-generated models shine. We cover the steps for bringing them into a game engine in How to Import GLB into Unity: Why It Is Not Supported Out of the Box and How to Install glTFast.
There's also the idea of using it as a base for an avatar. You build a rough 3D form from your illustration in one go, then use it as a reference while you do your own modeling or retopology. Because you skip building the shape from scratch, you can iterate on the silhouette faster. We've laid out all the editing routes in How to Edit AI-Generated 3D Models: A Complete Overview of Taking Them into Blender, Game Engines and 3D Printing.
For rough costs of commissioning a 3D model you can move on stream as a VTuber, and where to draw the line on what a non-moving 3D model can do, see How Much Does a 3D VTuber Model Cost? Crowdfunding Examples and What a Non-Rigged 3D Model Can Do.
FAQ
Isn't there a tool that converts an AI-generated model into VRM with one click?
Exporting to the VRM format is technically possible, but being able to export it and having it work as an avatar in VRChat are separate matters. VRChat officially requires humanoid avatars to have bones mapped, including head, hands and feet, and a hierarchy with defined parent-child relationships. A model without bones and weights won't meet these requirements no matter which format you convert it to.
If I outsource the rigging, what should I hand over?
Things move faster if you provide the cleaned-up mesh (GLB or FBX), the intended use (VRChat or another platform), and whether you need expressions. For VRChat, sharing the official rig requirements page helps everyone stay on the same page. Also note that the drawing you start from must be an original character you hold the rights to. Turning someone else's copyrighted character into 3D and uploading it is not something you can do without the rights holder's permission.
If I use it as an object placed in a world, what should I check?
The polygon count and the texture resolution. Worlds load many objects at the same time, so one heavy model can affect the whole experience. Check the polygon count before importing, and reduce it if needed before placing the model. We explain the nature of the GLB format in What Is a GLB File? How to Open, Use, and Convert It, Explained Simply.
Summary
Turning an AI-generated 3D model into a VRChat avatar requires rigging and weight painting, a separate production stage that file format conversion cannot replace. VRChat's official requirements specify bone mapping including head, hands and feet, right down to the parent-child relationships of the shoulders, neck, arms and legs. As for VRM, UniVRM can load glTF / GLB, but avatar information isn't generated automatically from a model with no skeleton.
On the other hand, for uses like objects placed in a world, subjects for photos, display pieces and a base for sculpting, you can use the generated model as it is. "Far from being an avatar, but close for everything else": keeping this line in mind from the start lets you use these models without mismatched expectations. If you want to dive into avatar creation itself, the surest way to begin is to check the overall learning order in Can't Make a VRChat Avatar? A Beginner's Path, Sorted by Where You Get Stuck.




