Nine GLB organs make a focused explorer, not a full atlas
Anatomy Atelier contains 9 interactive models: heart, brain, lungs, liver, kidneys, eyeball, intestine, pancreas, and skin. Each organ has labelled hotspots with anatomical terms, a written overview, basic measurements, a medical note, and a short fact. Supporting cards show tissue, location, comparison, and organ artwork. That is enough for a guided introduction, but the name can suggest far broader body coverage than the repository provides.
The viewer itself has thoughtful controls. A learner can rotate and zoom the model, isolate structures, toggle a cross section or layers, reset the camera, and enable automatic rotation. Clicking a hotspot opens its label and detail. A quiz asks the learner to find each labelled structure once, then reports a score. These interactions make the models more useful than a passive spin-around object.
Twelve languages are built into the same organ interface
The source defines 12 locales: English, Spanish, Hindi, Chinese, Arabic, Portuguese, French, German, Japanese, Russian, Indonesian, and Korean. Arabic switches the interface to right-to-left layout, while the font loader groups writing systems. Organ facts and interface text have separate dictionaries, which is a sensible structure for translators working on medical labels without touching navigation copy.
Language breadth does not verify scientific accuracy. The application includes measurements, medical notes, conditions, and Latin terminology, yet the repository does not identify an editorial or clinical review process. Treat the text as learning material that needs checking, especially before classroom or patient use. The same caution applies to hotspot positions: issue 27 shows heart labels such as the aorta offset from the displayed structure.
What happened when we ran it
Our sandbox installed commit 8c0e6f3 in 35 seconds. Npm added 773 packages, and the installed environment occupied 1,075 MB. The source checkout contained 131 files, about 6,874 lines, and used 35.3 MB. A vinext build then completed in 30 seconds inside an unprivileged Debian container with 3 CPUs, 8 GB of RAM, Node 22, and no secrets.
The test command failed after 10 seconds. Node's runner reported 2 tests, 0 passes, and 2 failures, with no skipped or cancelled cases. The supplied log tail contains the runner summary and a stack frame from pending subtests, but not the names or assertion messages. We cannot assign a cause from that evidence. The measured commit therefore builds while offering no passing automated test in the command we ran.
Npm audit found 32 known vulnerabilities: 1 critical, 24 high, 6 moderate, and 1 low. The lab also found no CI workflow file and no Dockerfile, though a tests directory exists. A successful 30-second build is useful, but it does not answer whether the viewer behaves correctly across browsers, whether the medical labels are accurate, or whether the dependency findings reach the deployed bundle.
The README documents vinext instead of Anatomy Atelier
The 3,439-character README calls the project vinext-starter. It explains Node 22.13, optional Cloudflare D1 and Drizzle files, workspace identity headers, and ChatGPT sign-in helpers. It never describes the nine anatomy models, the quiz, the 12 languages, the Three.js viewer, or the medical content. A developer can find npm install, npm run dev, and npm run build, but not the product they are about to maintain.
Starter residue remains in the codebase too. The package name is site-creator-vinext-starter, an empty database schema and D1 example are checked in, and the README calls those optional. In the rendered interface, Systems, Library, and Notes appear in the main navigation without click handlers in the component we inspected. The anatomy work is real; the repository packaging has not caught up with it.
Missing model provenance blocks safe reuse
The repository API detects no license, and there is no root license file. Issue 23 asks for reuse terms for the models and illustrations, while issue 24 asks whether the GLB files were created by the author or sourced from another dataset. Neither issue has a documented provenance answer. Publicly readable source is not the same as permission to copy, modify, or sell the bundled assets.
That gap changes the recommendation for developers more than a rough button does. You can inspect the Three.js loading, hotspot, disposal, quiz, and localization code. You should not assume you can carry the 9 models or their illustrations into another product. Human Atlas or the BodyParts3D files are clearer starting points when a documented source dataset is part of the requirement.
An August 9 push and 27 open items signal a young project
GitHub recorded the last code push on August 9, 2026, one week after the repository was created, and listed 27 open issues and pull requests combined. The API returned no latest release. Issue discussion continued into September, including licensing, model provenance, hotspot placement, and a report about a separate custom domain. The project attracted more than 3,400 stars quickly, but its release and maintenance history is still short.
Anatomy Atelier is easy to admire as an interface study. Nine models, 12 languages, label quizzes, and WebGL controls make it unusually complete for a small visual project. The decision changes once accuracy, redistribution, or maintenance enters the room: 2 failed tests, 32 audit findings, no asset license, and an unrelated README leave too much for a school or product team to resolve on its own.