MMD to VRM converter — Japanese bones read as they are
VTubeMe's MMD to VRM converter turns an exported MMD model into a VTubing-ready VRM online — Japanese bone names mapped to the VRM humanoid, MToon materials, 3D preview before download. センター, 上半身, 腕 and the rest are read exactly as the model author wrote them, in both the raw MMD spelling and the Blenderized one. One caveat first, and it is the important one: the converter does not read .pmx.
PMX is not read — export to FBX or glTF first
The converter accepts FBX, GLB, glTF, a Blender .blend project and a ZIP of a glTF folder. A .pmx or .pmd is none of those, and uploading one will not work. This page says so plainly rather than letting you spend a credit finding out.
The usual way out of PMX is Blender with the mmd_tools add-on: import the .pmx there and export the model as FBX or glTF. Then zip the export together with the model's texture folder — MMD distributions keep their images in a tex/ folder next to the model, and a model file only points at textures by filename, so the images have to travel with it — and convert the .zip.
The bone names survive that round trip, which is the part that matters here. Both shapes are recognized: raw MMD's side prefix (左腕) and the .L/.R suffix mmd_tools writes when it Blenderizes a rig so Blender's mirror tools work (腕.L).
Why 下半身 is not the hips
In an MMD skeleton 上半身 and 下半身 are siblings under センター, not a chain. So the VRM hips slot goes to センター, and 下半身 is left out of the humanoid on purpose: giving it the hips would put the spine outside the hips subtree and break the VRM hierarchy the apps rely on.
It is worth knowing because the converter's report lists unmatched bones, and 下半身 will be among them on a correctly converted model. That is the intended outcome, not a missing bone.
MMD bone names the converter maps
Left side shown; the right side is the same word with 右 instead of 左 (or the .R suffix). Where a joint has two common writings, both resolve to the same VRM bone.
| MMD bone | VRM humanoid bone |
|---|---|
センター(centre) | hips |
上半身(jōhanshin, upper body) | spine |
上半身2(jōhanshin 2) | chest |
首(kubi, neck) | neck |
頭(atama, head) | head |
左腕(hidari ude, left arm) | leftUpperArm |
左ひじ / 左肘(hiji, elbow) | leftLowerArm |
左手首(tekubi, wrist) | leftHand |
左足(ashi, leg) | leftUpperLeg |
左ひざ / 左膝(hiza, knee) | leftLowerLeg |
左足首(ashikubi, ankle) | leftFoot |
左つま先(tsumasaki, toe) | leftToes |
下半身(kahanshin, lower body) | left unmapped on purpose — see above |
左肩 and 右肩 take the shoulder slots, 目 the eyes, and the finger chains are read too: 親指 thumb, 人差指 (also written 人指) index, 中指 middle, 薬指 ring, 小指 little, numbered outwards from the hand. Both the hiragana and kanji writings of the elbow and knee resolve, and so do both side conventions, so a model that went through mmd_tools converts the same as one that did not. Verified on real files in our regression corpus. On a successful conversion the result panel names what it recognized: “mmd rig”.
What can go wrong with a MMD file
A conversion that cannot finish tells you which of these it hit, and returns the credit — you are never charged for a failure.
PMX_IN_ZIP
The archive holds a .pmx (or .pmd) and no .gltf, .glb, .fbx or .blend — the usual cause is zipping the MMD model itself.
Export the model to FBX or glTF first, then zip that export together with its tex/ folder.
NO_TEXTURES
The rig converted, but every texture the model asks for is missing, so it would come out grey.
Zip the exported model together with the tex/ folder and convert the .zip. Your credit is returned.
TEXTURES_UNREFERENCED
The upload does contain images, but the model file never references any of them, so there is no way to tell which texture belongs to which part.
Rename each image after the material it should cover — the failure message lists the material names — or send the files straight from the result panel.
RAR_IN_ZIP
RAR archives are unpacked here, but this one would not open — usually password-protected, damaged, or one part of a multi-part set.
Unpack the .rar on your computer and re-pack the model folder as a plain .zip.
Frequently asked questions
Can I convert a PMX file to VRM?
Not directly — the converter reads FBX, GLB, glTF, .blend and ZIPs of a glTF folder, and .pmx is none of those. The route that works: open the .pmx in Blender with the mmd_tools add-on, export the model as FBX or glTF, zip it with its tex/ folder and convert that. Nothing is lost in the detour as far as this converter is concerned — the Japanese bone names come through unchanged and are recognized as they are.
Does the converter understand Japanese bone names?
Yes, as-is: センター becomes the VRM hips, 上半身 the spine, 上半身2 the chest, 首 and 頭 the neck and head, 腕 / ひじ / 手首 the arm chain and 足 / ひざ / 足首 / つま先 the leg chain, plus all five finger chains. Elbows and knees written in kanji (肘, 膝) resolve the same as the hiragana forms, and the .L/.R suffixes mmd_tools produces are read alongside the 左/右 prefixes. There is nothing to rename before uploading.
Why do MMD models come out without textures?
Because the model file does not contain them — it points at them by filename, and MMD distributions keep the images in a tex/ folder next to the model. If only the exported model is uploaded, every texture it asks for is missing and the conversion stops with NO_TEXTURES rather than handing you a grey avatar; the credit comes back. Zip the exported model together with the tex/ folder and convert the .zip.
Is 下半身 missing from my converted model?
It is left out on purpose. In MMD, 上半身 and 下半身 sit side by side under センター rather than in one chain, so the VRM hips has to be センター — mapping it to 下半身 would leave the spine outside the hips subtree and break the hierarchy VRM apps expect. Seeing 下半身 in the unmatched list is the correct outcome.
Will the converted model work in VSeeFace?
The converter writes VRM 0.x, which is the version VSeeFace reads, so it loads directly. If your target app is on the other side of the split — Desktop Mate's Custom Avatar Loader wants VRM 1.0 — the version converter rewrites the file either way, keeping bones, expressions, look-at, spring bones and MToon materials. Open the VRM 1.0 ↔ 0.x converter →
What does it cost?
The first conversion is free. After that, credits, and a credit is only spent on a successful conversion: failures refund automatically, and a result that looks wrong in the 3D preview can be refunded self-serve before you download it. Inspect the model in the preview above before spending the download.
Mixamo character instead?
mixamorig: bones are mapped, and motion clips get their own mode.
Character Creator, AccuRig or ActorCore?
CC_Base_ bones are mapped and limb twist bones folded in.
Every other input format
The same converter takes FBX, GLB, glTF, .blend and Sketchfab-style ZIPs.
No 3D model at all?
Create a photorealistic VRM avatar from a single selfie instead.