A toggle is any part of a VTuber model that a hotkey can switch on, off or swap while you are live: an accessory, an outfit variant, an eye texture or a facial morph. Here is how the four common kinds work under the hood and how to wire one up without breaking the rig.
A toggle is one state change on your avatar that a hotkey, a Stream Deck button or an API call can trigger while you are on air. Glasses appear. The jacket disappears. The eyes change colour. The word comes from the Live2D and VTube Studio side of VTubing, where a hotkey list is the normal way to switch model parts, but VRM avatars run the same idea under other names: blendshape weights, object visibility and material swaps.
Most people say toggle when they mean three decisions bundled together: which geometry changes, what changes about it, and what fires the change. Separate those three and the rest is configuration.
This article answers the four questions that come up first: what counts as a toggle, which kinds exist, how each works under the hood, and how to build one in VTube Studio or a VRM pipeline without wrecking the rig or the draw call budget.
Two conditions separate a toggle from a variant: the change has to be reversible at runtime, and it has to be bound to a trigger you control. A jacket you remove by reloading the model is a variant. A jacket that vanishes when you press F7 and returns when you press it again is a toggle. Reversibility is the dividing line, and it is why a switch has to be planned while the model is built rather than patched in afterwards.
The trigger is usually a keyboard hotkey, but the same state change can come from an expression hotkey, a Stream Deck button, a chat command routed through the runtime API, or a face-tracking parameter. VTube Studio ships a hotkey system and an API, and most 3D runtimes expose something similar. It only has to be instant and repeatable.
| Toggle kind | What changes at runtime | Where it is authored | Typical trigger |
|---|---|---|---|
| Visibility | A mesh or ArtMesh stops being drawn | Separate object or ArtMesh in the model file | Hotkey that flips visibility or opacity |
| Texture and material | A material slot points at a different texture | Material slots in the DCC tool | Hotkey that swaps the material |
| Morph and expression | A blendshape weight moves from 0 to 1 | Blender shape keys or VRM blendshape groups | Expression hotkey or tracking parameter |
| Physics and parameter | A physics rule or parameter is overridden | Runtime physics settings, not the mesh | Hotkey that fixes a parameter value |
Hiding geometry is the cheapest runtime change there is: the renderer skips the object, so it saves work instead of adding it. The catch is that visibility is decided per draw unit. If a hat and the hair sit in one skinned mesh with one material, no flag can hide half of it. Keep every switchable piece as its own object in Blender or ArtMesh in Live2D Cubism, because splitting a rigged mesh afterwards means redistributing weights and rechecking UVs.
A texture swap costs no extra geometry work, but every variant has to be resident in memory or loaded on demand. A 2048 by 2048 RGBA texture is roughly 16 MB decoded before mipmaps, so five variants of the same atlas can add close to 80 MB. Artists avoid that by sharing one atlas and repainting only a small UV region, or by dropping small pieces to 1024 by 1024.
Three mechanisms cover almost every toggle. The first is an activity flag: the renderer skips the object entirely, the cheapest option with any mesh. The second is an opacity value: the object still renders and still costs fill rate, and overlapping transparent pieces can sort incorrectly and flicker. The third is a morph target weight, a number from 0 to 1 that pushes vertices along a stored offset, which is how facial expressions and shape changes such as shrinking an accessory are stored.
For a Live2D model the route is short. Split the outfit or accessory into its own ArtMesh in Live2D Cubism, export the model, then open the hotkey settings and create a hotkey whose action toggles that ArtMesh, normally by driving its opacity between 0 and 100. A motion toggle binds a hotkey to an animation instead of a parameter, and an expression toggle plays an expression file that holds parameter values. Menu layout and option names shift between releases, so match the current version rather than an older tutorial screenshot.
A VRM avatar keeps the same idea in the skeleton. In Blender you either leave the accessory as its own skinned object and let the runtime hide it, or fold it into shape keys and drive those weights; UniVRM exports both as blendshape groups. VSeeFace, Warudo, VNyan and Virtual Motion Capture each bind hotkeys their own way, and each project stores bindings differently, so moving a model to another runtime usually means rebuilding the triggers by hand. Test with tracking running: a shape key that fights a tracking parameter will lose.
Model formats, hotkey systems and runtime features change between releases, and many commissioned models come with terms that limit editing or redistribution. Back up the source files before you split geometry, check the documentation for your runtime, and read the artist terms before adding a toggle to a model you did not build.
| Carrier | What it keeps for toggles | What it drops or complicates |
|---|---|---|
| GLB and glTF | Mesh nodes, material slots and named morph targets | Morph names outside the exported set, and unbaked rig logic |
| VRM | glTF plus a humanoid rig, blendshape groups and expression presets | Constraints, drivers and helper bones that were not exported |
| FBX | Skinned meshes, material slots, shape keys and animation clips | Clean interchange between unrelated pipelines; materials rarely survive |
| OBJ | Geometry and UVs only | Morph targets, skinning, and materials beyond a companion MTL file |
| Live2D model3.json | ArtMeshes, parameters and expressions used by VTube Studio hotkeys | Anything outside the Cubism parameter and part model |
The consequence is that a toggle is only as portable as the format carrying it. Morph targets survive GLB, glTF and FBX, which makes shape keys safer for a model that may move between runtimes; object visibility works anywhere because it is a runtime decision, not file data. Two failures show up most often on stream: a switchable mesh sharing a texture atlas with a piece that must stay visible, and a hidden accessory still carrying a physics collider, so invisible hair keeps pulling on visible hair.
The toggle itself is rig work, and rigging is not something Open3DAI does: the generator produces meshes and textures, it does not author blendshapes, bind a model to a skeleton or add a hotkey. What it can supply is the parts. Generate glasses, a hat, a charm or a prop from a prompt or a reference image, compare engines if you want options, then export GLB, glTF, OBJ, STL, PLY or 3MF into Blender to fit against the head bone and weight it. AI texturing covers the material side. UV work, retopology and the hotkey binding still happen in Blender and in the streaming runtime.
No. An expression is a stored set of parameter values, usually for the face. A toggle is any state change a hotkey can flip, including hiding a mesh or swapping a texture. Expressions are one mechanism a toggle can use.
Sometimes. If the piece is already a separate object or ArtMesh, binding a hotkey takes minutes. If it was merged into a larger skinned mesh, you have to split the geometry and recheck weights and UVs, which is real work and often easier to redo from the source file.
There is no fixed limit. The practical limits are your hotkey layout, texture memory and how many states you can remember mid-stream. Typical commissioned models ship with a handful of outfit or accessory switches, and heavy users stack twenty or more across layered hotkey sets.
Visibility toggles usually help, because a hidden object stops being drawn. Texture swaps cost memory rather than frame time. Morph targets add a small amount of vertex work while the weight animates.
The runtime is applying a default state from the model file or from its own settings. Define that default deliberately, and if the runtime lets you save a startup state, do it there rather than relying on memory.
Yes, with different mechanics. Live2D models toggle ArtMeshes, parameters and expressions. VRM models toggle object visibility, material swaps and blendshape weights. Which is easier depends on the runtime you stream with.
It can make the part: the accessory mesh and its textures, exported as GLB or OBJ. Blendshape authoring, skinning and hotkey binding happen in Blender and in the runtime. Open3DAI states plainly that it does not rig or animate models.
It depends on the artist and the complexity. Simple visibility switches are often included in the base price, while a full outfit swap with new textures is typically quoted as an add-on. Rates change, so ask for a current quote instead of trusting a number in a blog post.
Send the same prompt or image to several engines, compare the meshes side by side, then export GLB, glTF, OBJ, STL, PLY or 3MF. Creating an account grants free credits and needs no card.
Surface Material Override in the editor, runtime swaps from GDScript, and the shared-resource behaviour that repaints every copy of a model — plus the PBR values and import settings that make the result hold up.
Cry of Fear runs on GoldSrc, so Simon is a compiled .mdl with a fixed skeleton and indexed BMP textures. Here are the two ways to change him, all twelve outfits, and what model packs really replace.
Texturing is four decisions taken in order: whether the model needs UVs, how the UV islands are laid out, which PBR maps you author, and where the pixels come from. Each decision is explained here with the numbers that make it.