mat-vis-client

Pure Python client for mat-vis — PBR texture data distributed as Parquet files on GitHub Releases.

• Zero dependencies — stdlib only (urllib). No pyarrow, no numpy, no binary wheels.
• HTTP range reads — fetch a single texture channel with one ranged GET against a consolidated Parquet asset. No whole-archive downloads.
• Local cache — results cached under ~/.cache/mat-vis/ with a configurable soft cap.
• Search — filter materials by category, roughness / metalness ranges, source, tier.
• MaterialX export — synthesized UsdPreviewSurface documents or original upstream authored documents (gpuopen).
• Format adapters — three.js, glTF 2.0 KHR_materials_pbrSpecularGlossiness, MaterialX.

Install

pip install mat-vis-client

Python 3.10+.

Quick start

from mat_vis_client import MatVisClient

client = MatVisClient()  # latest release

# Ergonomic path — VisAsset binds (source, material_id, tier) and
# exposes lazy scalars/textures + format adapters.
asset = client.asset("ambientcg", "Rock064", tier="1k")
three  = asset.to_threejs()   # MeshPhysicalMaterial dict
gltf   = asset.to_gltf()      # glTF 2.0 material dict
mtlx   = asset.to_mtlx()      # MtlxSource (.xml() / .export(dir))

# Lower-level — when you just want the bytes for one channel.
png = client.fetch_texture("ambientcg", "Rock064", "color", tier="1k")
with open("rock.png", "wb") as f:
    f.write(png)

Search

results = client.search(
    category="stone",
    roughness_range=(0.4, 0.9),
    source="ambientcg",
)
for r in results:
    print(r["id"], r["category"], r["roughness"])

Categories come from the release manifest (metal, wood, stone, fabric, plastic, concrete, ceramic, glass, organic, other). Inspect them at runtime with client.categories().

MaterialX

Synthesized UsdPreviewSurface wrapper — always available:

src = client.mtlx("ambientcg", "Rock064", tier="1k")
src.export("./out")        # writes .mtlx + referenced PNGs
xml = src.xml              # raw MaterialX XML

Original upstream document (gpuopen materials only; None elsewhere):

orig = client.mtlx("gpuopen", "<uuid>").original
if orig is not None:
    orig.export("./out")

Bulk prefetch

n = client.prefetch("ambientcg", tier="1k")  # download + cache every material

CLI

mat-vis-client list                              # sources × tiers
mat-vis-client materials ambientcg 1k            # IDs
mat-vis-client fetch ambientcg Rock064 color 1k -o rock.png
mat-vis-client search metal --roughness 0.2:0.6
mat-vis-client prefetch ambientcg 1k

Environment variables

Rate limits

Backed by GitHub's unauthenticated rate limit (60 req/h). The client:

• Caches signed CDN URLs for 4 min to skip the github.com redirect
• Retries 429 / 503 / rate-limited 403 with Retry-After / X-RateLimit-Reset honoring
• Raises mat_vis_client.RateLimitError when retries are exhausted

Adapters

The recommended entry point is client.asset(...).to_threejs() (see Quick start). The free functions below are the lower-level primitive layer — use them when you've already got scalars and textures dicts in hand (e.g. building a custom pipeline, porting to another language, or unit-testing without instantiating a client).

from mat_vis_client import to_threejs, to_gltf, export_mtlx

scalars = {"roughness": 0.5, "metalness": 0.0}
textures = client.fetch_all_textures("ambientcg", "Rock064", tier="1k")

three = to_threejs(scalars, textures)
gltf = to_gltf(scalars, textures)

Links

• Source / issue tracker: https://github.com/MorePET/mat-vis
• Changelog: https://github.com/MorePET/mat-vis/blob/main/CHANGELOG.md
• Data license: CC0 (AmbientCG, PolyHaven) + MIT (GPUOpen) per material