Start with a correction. Yesterday we reported that nvidia/simready-assets held a quarter of a million downloads with no licence. The repository does indeed declare no licence. We should have looked inside before saying there wasn’t one.
There is. It is CC BY 4.0, and there are 2,575 copies of it — one per asset, in <asset>/.metadata/LICENSE.txt, as the complete, unabridged Creative Commons Attribution 4.0 International legal text.
That changes the story from “NVIDIA published unlicensed assets” to something more interesting: the most permissively licensed industrial 3D library on Hugging Face is invisible to anyone who searches by licence.
What is actually there
The dataset is the Hugging Face half of SimReady Central, announced on 21 October 2026. In the day to 26 October it gained 138,003 downloads (255,193 → 393,196, up 54.1%) and 210 likes (134 → 344, up 156.7%), making it the biggest dataset likes gain on our tracker and the second-biggest download gain.
We paginated the repository tree rather than trusting the file listing, because the Hugging Face API caps siblings at 100,000 entries and this repository exceeds it. The real figure:
- 2,575 asset directories (the API’s capped listing shows only 1,663 of them)
- Formats across the sampled portion: 21,524
.usda, 3,362.usd, 5,312.mdlmaterial definitions, 59,717 PNG thumbnails and textures, 8,369 JSON metadata files
Secondary coverage says “more than 2,000 assets”. It is 2,575.
The category mix is the part nobody has mentioned, and it is the reason this matters to artists rather than only to robotics teams. By directory-name prefix, the largest categories are:
| Category | Assets |
|---|---|
| food | 522 |
| kitchenware | 179 |
| box | 154 |
| handTool | 97 |
| consumable | 77 |
| paint | 68 |
| bodyPart | 67 |
| pallet | 64 |
| equipment | 59 |
| serviceTool | 57 |
| engineComponent | 52 |
The single largest category is food. Then kitchenware, then boxes. Alongside the V8 and V6 engine assemblies, spray booths, robot chargers and compressors you would expect from a factory-simulation library, there are 522 food props and 179 pieces of kitchenware — which is to say a free, commercially usable set dressing library for anyone who needs to fill a kitchen, a canteen or a shop shelf.
The paperwork, which is excellent
Each asset directory contains a .metadata/ folder. We read three of them in full.
com.nvidia.simready.packaging.json — the licence, machine-readable:
{
"format_version": "1.0",
"package_id": "simready.hf.nvidia.simready-assets.food_beverage_beerBottle_a01.2026.10.13_180542048590",
"license": "CC-BY-4.0",
"content_hash": { "sha256": "e3f46a6c25e2fe58b90d902ccc9a3f029298e476b1a84e9167fc270d5d93176d" },
"package_hash": { "sha256": "37515e1da7c6f897218c0397d8fb8da366ad26149fc760bb452f2f15b37e6cbc" }
}
A versioned format, a package identifier carrying the packaging timestamp, an SPDX-style licence string, and two SHA-256 hashes — one for the content, one for the package.
com.nvidia.simready.packaging.bom.json — a bill of materials listing every file with its relative_path, size and SHA-256. For one beer bottle that covers the USD, four thumbnail variants at different sizes and lighting conditions, and the textures.
com.nvidia.simready.conformance.Package@1.0.0.json — a conformance run:
{
"format_version": "1.0",
"profile": "Package",
"profile_version": "1.0.0",
"timestamp": "2026-10-13T18:05:42",
"assets": [ { "asset": "com.nvidia.simready.packaging.json",
"features": [ { "FET_030_STANDARD": { "version": "0.1.0", "passed": true,
"failing_requirements": [], "dependencies": [] } }, … ] } ]
}
A dated validation result against a versioned profile, with named requirements, explicit passed flags, an empty failing_requirements list, and declared dependencies between checks.
That is better provenance than the Scenario dataset we made yesterday’s lead, which was already the best-documented creative dataset we had read this month. Scenario gives you per-file hashes and a published validation summary. NVIDIA gives you per-file hashes, a per-asset licence, and a machine-readable conformance record with versioned requirement identifiers.
We checked coverage rather than assuming it. Within the 100,000-file listing the API returns, there are 1,663 asset directories and 1,663 LICENSE.txt files — zero assets without one. We then sampled five assets from beyond that cap, including three beer bottles and two bowls. All five: Attribution 4.0 International.
And the one line that is missing
Here is the whole repository root:
.gitattributes
That is it. No README.md. No LICENSE. The API returns cardData: null. The tags are size_categories:n<1K, modality:image, region:us — and the size category is wrong by three orders of magnitude, since the repository holds well over 100,000 files.
So on Hugging Face, this dataset has no card and no licence. Filter the datasets list by CC BY 4.0 and it does not appear. Land on the page and there is nothing to read. A studio’s legal review, looking at a repository with no stated licence, would correctly conclude that nothing has been granted — and would be wrong by the width of the most permissive licence NVIDIA could have chosen.
The irony is sharpened by the announcement itself. Hugging Face’s write-up of SimReady Central tells you, correctly, how to choose an asset:
A request for a ‘warehouse gripper’ is incomplete without payload needs, mounting compatibility, articulation requirements, runtime support, licensing, and the profile needed for the task.
Good advice. You cannot follow it on the platform page, because the licensing is the one field the platform does not have.
The companion repository makes it worse
There is a second repository, nvidia/simready-central-library, created 29 September and updated 21 October. It is the one named after the announcement. It has:
- a README
- a declared
license:apache-2.0tag - two files
- 123 downloads
- and it is gated behind manual approval
So the repository with the licence metadata, the card and the matching name is the one with nothing in it and a door on it. The repository with 2,575 assets and 393,196 downloads is ungated, undocumented and blank. Coverage describing SimReady Central as a “free, ungated starting point” is right about the assets and would mislead anyone who clicked the obvious link.
What this means if you want to use them
You can. CC BY 4.0 permits commercial use and modification with attribution, and the per-asset licence files are unambiguous about which licence applies.
Two practical cautions.
Attribution is a condition, and the licence files do not name a licensor. The CC BY text in each asset is the bare, unfilled Creative Commons template — unlike Scenario’s, which states “Attribution: Scenario” in its own LICENSE file. CC BY section 3(a) requires you to retain creator identification “if it is supplied by the Licensor”; where it is not supplied, you are left to work out what reasonable attribution looks like. Given that the library includes OEM-contributed models, crediting “NVIDIA SimReady Central” and linking the dataset is the defensible reading, and recording the package_id of each asset you ship is cheap insurance — it is right there in the JSON.
Check the profile, not the preview. The announcement is firm on this and it is correct: a SimReady profile is “a capability declaration, not a universal quality badge”. An asset validated for spatial planning need not carry the mass, friction or joint data a physics simulation wants. For set dressing none of that matters. For anything that touches or collides, read the conformance file before you trust the geometry.
And verify the hashes. They gave you a SHA-256 for every file. Almost nobody will check them. They cost nothing to check.
Sources
- nvidia/simready-assets — Hugging Face
- nvidia/simready-central-library — Hugging Face
- Simulation-Ready Assets Turn 3D Libraries Into Testable Robotics Infrastructure — Hugging Face Blog
- CC BY 4.0 licence deed
- OpenUSD