GLB to VRM API — three calls from a .glb to a talking VRM 1.0
Upload a humanoid .glb, rigged or not, and download a VRM 1.0 with the bones mapped and a face that talks and blinks. No Unity, no Blender, no person in the loop.
To convert a GLB to VRM programmatically, POST the .glb (up to 100 MB) to https://app.riggle3d.com/api/v1/conversions with your API key, poll GET /conversions/{id} every 3-5 seconds, and download GET /conversions/{id}/vrm when status is succeeded. The result is a VRM 1.0 with humanoid bones, five visemes, blinks, three emotions, and lookAt when the rig has eye bones, built automatically in 1-3 minutes even when the GLB has no skeleton and no face shapes. It costs 10 API tokens, $1.00 to $1.50 depending on the pack, and a failed conversion returns its tokens.
The three calls
export RIGGLE_API_KEY=rgl_...
API=https://app.riggle3d.com/api/v1
# 1. start a conversion (10 API tokens)
id=$(curl -s -X POST $API/conversions \
-H "Authorization: Bearer $RIGGLE_API_KEY" \
-F file=@character.glb | jq -r .id)
# 2. poll until it is no longer processing
until [ "$(curl -s $API/conversions/$id \
-H "Authorization: Bearer $RIGGLE_API_KEY" | jq -r .status)" != processing ]
do sleep 4; done
# 3. download the VRM
curl -s -o character.vrm $API/conversions/$id/vrm \
-H "Authorization: Bearer $RIGGLE_API_KEY"That is the whole integration: create, poll, download. The fields, the conversion object and every error are in the API reference, and a zero-dependency Python and Node client is in github.com/riggle3d/glb-to-vrm.
While you poll, stage tells you where the conversion is: rigging (only when the GLB had no skeleton), marking_face, then building_vrm.
Why this is hard to build yourself
A VRM is a glTF with extensions, so writing the container is the easy part. What takes the time is everything the extensions describe:
- The humanoid bones. VRM needs to know which joint is
hipsand which isleftUpperArm, and every source names and arranges its bones differently: Mixamo, Tripo, VRoid, Character Creator, game rigs, rigs with no usable names at all. - VRM 1.0 expressions. The file needs
aa,blink,happyand the rest. A GLB from an AI generator or an auto-rigger almost never has a single face shape, so there is nothing to name. - Visemes need a mouth that opens. A model made with its mouth closed has no opening at all, and moving the whole lower face looks wrong.
- Checking it. An expression can exist in the file and move nothing on screen.
The API hands you the finished file instead: a VRM 1.0 with the bones in place, a face that talks, blinks and emotes, and a preview video to check it by - whether or not the GLB had a skeleton or any face shapes.
What you send and what you get
In: one .glb of a humanoid character, up to 100 MB, as the multipart field file. Rigged or unrigged. An optional name, and an optional Idempotency-Key header so a retried request is not charged twice.
Out: a VRM 1.0 with humanoid bones, the visemes aa ih ee oh ou, blink, blinkLeft, blinkRight, happy, angry, surprised, and lookAt when the rig has eye bones; a preview.mp4 that plays every expression; stills listed in preview_images; and warnings.
Read warnings. An empty list means nothing looked off. A warning means the avatar was built but you should look at the preview: the face was found with low confidence, or a viseme barely moves. The quality of the mouth depends on the model. A character whose mouth was modelled open gives the most natural mouth shapes; a closed mouth gives softer movement.
Pricing
| Pack | API tokens | Price | Per GLB converted |
|---|---|---|---|
| Starter | 100 | $15 | $1.50 |
| Builder | 500 | $60 | $1.20 |
| Scale | 2,500 | $250 | $1.00 |
One GLB to VRM conversion is 10 API tokens. Polling, the VRM download, the previews and deleting are free. Tokens do not expire and a failed conversion returns them automatically. There are no free API tokens: a key works once a pack is bought.
Who it is for
- AI avatar and companion apps. Your users bring or generate a character; your app needs it to lip sync to TTS. Send the GLB, store the VRM, drive
aa ih ee oh oufrom audio. - VTuber tools. Accept any humanoid GLB at import and hand the user an avatar that already blinks and talks.
- Game and character pipelines. Batch-convert a library of characters to VRM 1.0, up to three at a time per account.
- 3D generators. You already produce a GLB. One request adds an "Export as VRM" button next to it.
Converting one model by hand? The GLB to VRM converter does the same job in the browser. Starting from a picture rather than a model? See the image to VRM API. Already have a VRM whose face does not move? See adding visemes and blinks via the API.
Convert your first GLB through the API
Make a key and buy a pack: 10 API tokens a model, returned if the conversion fails.
Get an API keyFAQ
Does the GLB have to be rigged?
No. A GLB with no skeleton works too; Riggle adds one, and while it does, stage reads rigging. The price is the same 10 API tokens.
My GLB has no blendshapes. Will the VRM still talk?
Yes, that is the case the API is built for. The VRM comes back with visemes, blinks and emotions, so the GLB needs no face shapes of its own.
How long does a conversion take?
One to three minutes for a rigged model. Poll GET /conversions/{id} every 3-5 seconds; there are no webhooks.
How much does it cost per model?
10 API tokens: $1.50 with the Starter pack (100 tokens for $15), $1.20 with Builder (500 for $60), $1.00 with Scale (2,500 for $250). A failed conversion returns its tokens.
Can I send a .gltf, an FBX or a URL?
No. The API takes a single .glb or .vrm uploaded in the request. Pack a .gltf with its .bin and textures into one .glb, and export an FBX to .glb, before sending.
Does it output VRM 0.x?
No, VRM 1.0 only.
What if the conversion fails?
The conversion's status becomes failed with error.code conversion_failed and a message, and its 10 tokens are returned automatically (refunded is true).
Can I convert many GLBs at once?
Up to three conversions can be in flight per account. Queue the rest on your side and start the next one when one finishes; a fourth request gets 429 too_many_active.
Is there a client library?
Not as a package. github.com/riggle3d/glb-to-vrm has a zero-dependency Python client and a zero-dependency Node client that you copy into your project, plus curl examples.