SLMocapArchive

Collected Sign Language Motion Capture

Hello, this is Studio Galt. Since 2020 we have been recording motion capture, primarily of American Sign Language (ASL). Everything here is free under CC0.

Browse the signs Search the archive, watch each sign, and download its files without cloning anything.

At a glance: about 2,700 motions covering 2,100+ signs, plus the ASL fingerspelling alphabet and numbers. Each motion comes as FBX for Blender, Unreal and Mixamo-style rigs (and loads in Unity and Maya), with a rendered video and a ReadMe. See catalog/SUMMARY.md for current counts and coverage.

1. Quick start
2. Finding a sign
3. Downloading only what you need (or a whole pack)
4. What's in a motion folder
5. Reporting an error or requesting a sign
6. General FAQ
7. Tech FAQ
8. Avatar FAQ
9. File FAQ

Quick start

1. Find your sign in the gallery, or in the catalog (use the row whose status is current).
2. Download just that folder (how).
3. Pick the FBX for your target:

I'm using…	Use this file
Blender, with the Galtis avatar	… Mesh.fbx (or the .blend in Blend File)
Blender Rigify	… Rigify APose.fbx / … Rigify TPose.fbx
Unreal Engine 5	… Manny / … Quinn / … MetaHuman FBX (newer motions) or … No Mesh UE Mannequin.fbx (older)
Mixamo-compatible rigs, Unity	… No Mesh Mixamo.fbx or … Mixamo TPose.fbx
Research / your own retargeting	… No Mesh Full.fbx (every bone), or the raw take in FBX Files/Original or Raw Mocap

Not every motion has every file. The file set changed over the years (see Folder layouts), and the catalog lists exactly what each motion has.

Finding a sign

catalog/catalog.csv lists every motion, one row each. It opens in any spreadsheet, and GitHub lets you search it in the browser. catalog/catalog.json has the same data for scripts.

Useful columns:

Column	Meaning
name, group	The sign as named in the folder; group gathers alternates (Alt 1, Var, 2, …) of the same sign
date	Upload date (ISO format, so it sorts correctly)
status	current, or superseded when the same motion was re-made under a newer date (superseded_by says which)
folder	Where it lives in the repo
layout	Which folder layout it uses
rig, blender_version	Galtis rig version and Blender version it was made with
raw_mocap_file	The raw take it was cleaned from, when known
raw_mocap_source	How that link was found: readme (the ReadMe's Original Motion), bundled (the copy in FBX Files/Original), or name match (a Raw Mocap file with the sign's name, used only when the ReadMe doesn't name a take; where both a plain and a _frames_ cut exist it is a best guess)
keypose_frames	Key pose frame numbers from the ReadMe
facial_animation	yes if face shape keys move beyond blinking, no if only blinks, unknown for motions whose ShapeKeys file has no key names (mostly 2023)
has_… columns	Which file types exist for that motion

The catalog is rebuilt automatically on every push by tools/build_catalog.py. Anything that looks misnamed or incomplete goes to catalog/issues.csv. A raw take that really was used for more than one sign is listed in tools/shared_takes.txt, so it isn't reported as a copied ReadMe value.

Downloading only what you need

The full archive is very large, so cloning everything isn't practical. Git can download just the folders you want:

git clone --filter=blob:none --sparse https://github.com/StudioGalt/Sign-Language-Mocap-Archive.git
cd Sign-Language-Mocap-Archive
git sparse-checkout set "SG ASL Dictionary/ASL C/SG ASL Cabinet 2025-2-12 Upload"

Copy the folder path from the folder column of the catalog. To add more folders later, run git sparse-checkout add "<folder>".

Without git: DownGit downloads a single folder as a zip. Paste in the folder's GitHub URL. There is a DownGit walkthrough in Documentation, and more guides (Maya, Unreal, Rokoko) in the Documentation index.

Download packs

Want every sign for one program? Each release has zip packs of all current signs, one file set per software. The same packs and the catalog are also on Hugging Face, for download with Hugging Face's tools (hf_hub_download, load_dataset):

Pack	Contents
unreal	UE5 Manny and Quinn, A-pose and T-pose (older signs: UE Mannequin)
metahuman	MetaHuman, A-pose and T-pose (signs from 2026)
mixamo	Mixamo skeleton, also for Unity Humanoid
rigify	Blender Rigify, A-pose and T-pose (signs from 2025)
blender	The full Galtis skeleton (newest signs: the Mesh FBX), plus the Galtis rig as a separate file

Big packs are split into parts by letter (e.g. SLMA-2026-11-11-blender-part1-A-C.zip; fingerspelling comes last, in a part ending -Fingerspelling). Each zip has a README and an index.csv of the signs inside, and SLMA-<release>-pack-index.csv on the release page says which zip holds each sign. Packs hold only the current version of each sign, with no videos or raw takes; for those, use the folders above.

What's in a motion folder

Each motion is a folder named SG ASL <Sign> <Date> Upload. The newest layout (2026) looks like this:

SG ASL <Sign> <Date> Upload/
├── Blend File/       Working.zip: the Blender file used to make the motion
├── Documentation/    ReadMe.txt, ShapeKeys.txt, Framelog.txt, captioned video (.webm)
├── FBX Files/
│   ├── Full/         Mesh.fbx: Galtis with mesh and every bone
│   ├── Mixamo/       Mixamo-compatible skeleton (most motions)
│   ├── Original/     the raw take before cleanup
│   ├── Rigify/       Rigify A-pose and T-pose
│   └── Unreal_Engine/ Manny, Quinn and MetaHuman, A-pose and T-pose
└── Poses/
    ├── CSV/          bone transforms for each key pose
    └── JSON/         the same poses for JSON Pose Manager (jpm)

Folder layouts

The layout has improved over time. Older folders are left as they were so existing links keep working.

Layout	Used	What's different
v1	2023 (and the 2023 fingerspelling)	All files loose in one folder: Mesh, No Mesh Full/Mini/Mixamo FBX, MKV and captioned MP4, ReadMe, ShapeKeys
v2	Nov 2023 – Apr 2024	Documentation and FBX Files subfolders; Poses added in 2024
v3	Apr 2024 – Mar 2025	FBX split into Game Ready (Mesh, Mixamo, UE Mannequin, Rigify) and Research (Full, Mini, UEPlus); pose JSON; GIF; Blend File from Jan 2025
v4	Mar 2025 – Jan 2026	FBX grouped by target: Full, Mixamo, Rigify, Unreal_Engine (Manny, Quinn), Other (Mini, UEPlus, until Dec 2025); the GIF is no longer included from mid-2025
v5	Jan 2026 – now	Adds FBX Files/Original (raw take), MetaHuman FBX and Poses/CSV; Mini, UEPlus and the pose .blend libraries are no longer included

Reporting an error or requesting a sign

If a sign is wrong or a file doesn't load, please open an issue using the Sign correction or File problem template. Corrections are re-recorded under a new date (General FAQ 12), and the catalog marks the old one as superseded.

Missing a sign? Use the Request a sign template, or add a 👍 to an existing request. The most-wanted signs are recorded first.

Every sign has its own page you can link to, for example Cabinet. Use Copy link in the gallery to get it.

If you use the archive in research, please cite it:

Studio Galt (2026). Sign Language Mocap Archive (Version 2026-11-12) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.24183343

GitHub's Cite this repository button gives the same citation in other formats, from CITATION.cff. The Zenodo record holds a permanent copy of the catalog, documentation, pipeline scripts and rig; the motions themselves are here on GitHub. Citing is appreciated, not required: everything is CC0.

---

GENERAL FAQ

1. What's the License?

CC0. We don't reserve any rights; use this however you like. We want to make these assets free for everyone.

2. CC0 forever?

Yes, unless we are forced by law to change. So please, no crimes.

3. Why Sign Language?

It hadn't been done, or wasn't publicly available. We mainly do ASL (American Sign Language).

4. Why ASL?

I'm in English-speaking North America.

5. Speaking of which, you're not Deaf?

Nope, I'm a hearing person, and my knowledge of ASL is not complete.

6. Why would I come to you for translation then?

You shouldn't. I do not offer translation services. I offer the ability to turn translation into animation. Once it's digital, a lot can be done quite easily: change the character, the angle, the speed. Digitization doesn't solve every problem, but it has its advantages.

7. I speak English, why don't I just do it myself?

That would be SEE (Signed Exact English), not ASL. That may or may not be appropriate, depending on your project. Like a dictionary, it can teach certain things about a language, but not others.

8. Do you intend to do other languages?

We try to do the base alphabets for as many languages as we can, but a full additional language would require sponsorship and help with the language.

9. Sponsorship?

This is not my full-time job, but I would like it to be. We will fight to keep the resources free and CC0 in any case.

10. What do you mean by language assistance?

Most sign language resources are written in the local spoken language. I can do ASL and BANZSL (British, Australian and New Zealand Sign Language) because the resources are in English, but working without knowing the spoken language would multiply the errors.

11. Errors?

While 100% accuracy is the goal, mistakes can and will happen. That's yet another reason we recommend professional translation. We do have a system for correcting errors.

12. How do you correct errors?

Every file has a date on it. When an error is found, we recreate the motion under a new date. We try to avoid deleting old motions, even incorrect ones, because we believe it's important to be able to reproduce any results (even the errors). Rule of thumb: newer is more accurate. The catalog's status column does this comparison for you.

13. Can I use the motions for non-sign language purposes?

Yes, but these motions are part of a language, so they have meaning, and using them out of context can confuse and annoy people (see Neon Genesis Evangelion). A good way around this is to modify them so they're less likely to be mistaken for something they aren't.

14. We can modify these files?

Absolutely. That is the point of this project. Retarget them to your characters.

15. Can I use this for sign languages other than ASL?

If you know a sign in another language matches an ASL one, or how to modify an existing motion to match, go for it. Please let us know too: if this project ever tackles other languages, being able to reuse existing motions would be a great help.

16. Can I use this for machine learning (AI)?

That's a big category, so let's break it into sub-points.

16a) AI translation?

ASL and English are different languages, so automatic translation will be difficult. Great automatic translation is definitely the long-term goal, but the timeline could be anywhere from tomorrow to never. There is definitely value in fast but dubious-quality translation; the popularity of fan-subbed anime around the globe is proof of this. If someone says automatic translations are not desirable for them, please respect their wishes.

16b) AI art?

For art AI, yes, but please respect other artists' decisions not to have their work used for AI art.

16c) Are you using AI?

Maybe. The Xsens and StretchSense systems have algorithms to smooth and correct data, and since I don't know their internals, I can't say for sure. Coding is AI-assisted, and there are plans for AI automation.

16d) AI automation how so?

The workflow now exports records of changes in layers as CSV. The idea is to use reinforcement learning to predict where edits will be needed, to speed up cleanup.

17. Can we use them for NFTs?

Yes, but we will not endorse any NFT projects. We are for maximum access for everyone, whereas NFTs are about controlled access. That's not a bad goal, but it's not ours.

TECH FAQ

1. What do you record at?

All recordings are done at 240 frames per second, but we post at 60 frames per second.

2. Why change fps (frames per second)?

To limit file size and reduce cleanup time.

3. What is your equipment?

An Xsens Link suit and StretchSense gloves.

4. Are you sponsored by them?

No, but I would like to be.

5. What software do you use?

Blender is the primary tool. It's free and open source, and one of the goals of this project is to create sign avatars at zero cost (or as close as possible), so Blender was the natural choice. We do occasional quality checks to make sure files load into Autodesk Maya, NVIDIA Omniverse, Unity and Unreal Engine, but not every motion is checked. If something doesn't work, please let us know!

6. What about X software?

We don't have access to every piece of software, but Blender is very good at interoperability. If there's software you'd like us to test with, open an issue.

7. What are your export formats?

FBX is our primary format. We want to expand into glTF and USD. USD is an open format created by Pixar and is being adopted across much of the 3D industry.

8. Facial data?

ASL and other sign languages use facial movements to convey tone and information. Some signs require facial movement to be complete; for others it adds extra information (similar to punctuation or vocal tone). Because there is a wide range of possible tones, facial data is usually left blank unless the sign requires it. The catalog's facial_animation column shows which motions have it.

9. How do I change her face?

Shape keys, sometimes called blendshapes. Select the GaltisHead object and open the Data tab (green triangle on the right). You will see a list of shape keys following FACS (the Facial Action Coding System). Move the values between 0 and 1 to see the face perform each movement.

10. Where are your tutorials?

See the Documentation index for workflows in Maya, Unreal (mannequin and Mixamo characters) and Rokoko. If you have something to share, or want something covered, let us know.

AVATAR FAQ

1. Who is the Avatar?

Her name is Galtis. Very creative. But we stan.

2. Why is she a girl?

Hatsune Miku demands it. The real reason: part of the cleanup process is making sure the mesh doesn't intersect itself, which is more likely to happen on a female avatar than a male one. If intersections are fixed for the female avatar, they should be fixed for a male avatar too.

3. Can I use her in X?

She is open source and free to use. Please, no crimes.

4. What is the rig flow?

The animator manipulates FK and IK bones, which move an intermediate rig. The intermediate rig has correction bones that help it deform more naturally. Finally, the correction bones and intermediate bones drive the deform rig that goes into Unreal. All the deform rig does is translate the axes from Blender's standard to Unreal's.

Rig versions are in the Rigs folder, named by date (Galtis 8 YYYYMMDD.blend). Each motion's ReadMe and the catalog's rig column say which version it used.

FILE FAQ

1. There is a lot going on in each folder.

Correct. Let's break it down. (Folder names vary by era; see Folder layouts.)

2. What is in the Documentation folder?

The MKV is a video of the rendered animation, and the MP4 (or, in newer motions, the WebM) is the same video with captions. That in turn becomes the GIF used on social media. The ShapeKeys text file shows which shape keys are active and when, so you can recreate the face movements. The ReadMe lists the Blender version, rig version, date, the original take, and all key pose frames. Newer motions also have a Framelog, which records any frame-range adjustments made during cleanup.

3. What is in the Poses folder?

The JSON folder has a JSON file for every key pose, listing all bone positions. These work with JSON Pose Manager (jpm) for Blender, and can be used for non-jpm purposes too.

Newer motions also have a CSV folder with the same key poses as plain spreadsheets, including Rigify metarig versions (metarig A / metarig T). The .blend files (in older layouts) contain nothing but the pose assets. You can append them to your own copy of the rig to access the poses. The Mixamo one is exactly the same, but with Mixamo bone mapping.

4. What is in the FBX folder?

It depends on the era (see Folder layouts), but the files are the same family:

File	What it is
Mesh.fbx	Galtis with mesh and every bone (almost 400). Probably overkill, but it's there.
No Mesh Full.fbx	Every bone, no mesh. Heavier, good for research.
No Mesh Mixamo.fbx / Mixamo TPose.fbx	Matches the Mixamo rig, plus a root locator.
No Mesh Mini.fbx	Very similar to Mixamo, with different bone names (in case Adobe ever asks us to stop mapping to Mixamo, we'll switch to Mini).
No Mesh UE Mannequin.fbx	Matches Unreal Engine's mannequin.
No Mesh UEPlus.fbx	UE Mannequin plus extra correction bones.
Manny / Quinn / MetaHuman APose / TPose	Unreal Engine 5 skeletons, in A-pose and T-pose versions.
Rigify APose / Rigify TPose	Blender Rigify skeleton.
FBX Files/Original/…	The raw take the motion was cleaned from.

5. Do I have to download the entire archive to get the file I want?

No. See Downloading only what you need.

6. What is the Raw Mocap folder?

The uncleaned recordings, in numbered batches. The file name is the take name plus its frame range, e.g. SG ASL Cabinet_2078-2645.fbx. Batches 3–6 also contain _frames_ re-cuts of many takes (e.g. SG ASL Black_frames_521-1294.fbx); either cut may have been used (checking the working files found the _frames_ cut behind about one in six motions where both exist). A motion's ReadMe ("Original Motion") and the catalog's raw_mocap_file column point to the take it came from.

7. What are the scripts in Documentation/Scripts?

The Blender scripts used to build each motion, from import through bundling (the 2026 versions, plus the July 2024 ones for reference). See Documentation/Scripts/README.md for set-up and order.