Mapshroom Deep Brush Lab — independent experimental implementation.

UI/workflow reference:
Adobe Project Deep Brush, MAX Sneaks November 2026.
https://labs.adobe.com/sneaks/deep-brush
https://blog.adobe.com/en/publish/2026/11/11/adobe-max-sneaks
Public demo screenshots were used as design references; no Adobe app code or unmodified screenshots are bundled. The parkour and draped-cloth examples are AI recreations based on user-supplied demo screenshots.

Depth estimation:
onnx-community/depth-anything-v2-small, Apache 2.0
https://huggingface.co/onnx-community/depth-anything-v2-small
Pinned model revision: 4472b7362082ad9968fee890ca0f1e5aca36b93d
Transformers.js 4.3.1, Apache 2.0 (vendor/TRANSFORMERS-LICENSE.txt)
ONNX Runtime Web 1.31.0-dev.20260914-8d85527a0, MIT (vendor/ONNX-LICENSE.txt)
Browser worker uses q8 and single-thread WASM. Model files are fetched from Hugging Face; photographs are not transmitted.

Geometry:
MapshroomV3/src/lib/depthNormalPacked.js reused unchanged from the user's Mapshroom workspace.
Same RG octahedral normals / B relative inverse depth / A coverage format.
Demo depth provenance: assets/depth/provenance.json (existing Mapshroom Colore Lab predictions).

Demo photos:
coffee.png — Rachel Michetti / Pikolo Espresso Bar, CC0, via scikit-image.
astronaut.png — NASA Eileen Collins portrait, public domain, via scikit-image.
brick.png — scikit-image example image.
https://scikit-image.org/docs/stable/api/skimage.data.html
Original example files sourced from scikit-image v0.25.2.
faraday.jpg — AndyScott, CC0 1.0.
https://commons.wikimedia.org/wiki/File:Royal_Institution,_bust_of_Michael_Faraday.jpg

Typography: DM Sans via Google Fonts; browser system fonts are the fallback.

GENERATED SURFACE STUDIES (2026-11-13)
Blue hatchback, inclined concrete wall and cream linen: built-in AI image generation; original generated compositions, not Adobe event photos. Matched depth via Depth Anything V2 Small; normals and packed maps via Mapshroom. Provenance and SHA-256 hashes accompany each set at assets/generated/{car,wall,fabric}/provenance.json.
Adobe demo references: https://labs.adobe.com/sneaks/deep-brush and https://video.tv.adobe.com/v/3492660/ (concrete, fabric and portrait demonstrations).

UPDATE — 13 November 2026
Added generated draped marble statue and terracotta vase with Depth Anything maps. The default car depth is now generated with OpenAI imagegen; normals and combined are derived from that generated depth. This is an approximate geometry interpretation, not calibrated metric depth. Per-image provenance.json identifies the actual depth source.

REFERENCE SCENE RECREATIONS — 13 November 2026
Parkour wall and draped cream cloth: generated with built-in OpenAI imagegen from user-provided Adobe demo screenshot references, removing the video/brush overlays. These are generated recreations, not downloaded original source photos. Depth Anything V2 Small estimates depth from each final photo; Mapshroom derives normal and combined maps. Full provenance and hashes are at assets/generated/{parkour,cloth}/provenance.json. Existing gallery samples are retained.