Claude-Built X-Ray Mirror Adds 3D Anatomy to Your Webcam, but It Can't See Your Bones

What it is​

Reddit user Sourcecode12 posted the tool on the ClaudeAI subreddit. It layers an "x-ray" view of your bones over a webcam feed. The GitHub repository describes it as a live webcam X-ray mirror, built with Claude Fable 5.2.

The timing matters. Anthropic released Claude Fable 5.2 on November 9, 2026, so this demo appeared within days of launch. A community list of Fable 5.2 creations says X filled up with games, 3D worlds and web pages made with it within days. It also lists the X-Ray Mirror among them.

Social posts credit a few different people with the demo, including a post from Hashem Al-Ghaili dated Nov 10. I can't settle the authorship question here. The XDA report and the maker's Reddit post point to Sourcecode12, while the demo is hosted under the GitHub account MasterOfVFX2070.

What it does​

The page tracks your face in real time and fits a 3D anatomy model to it. The posts say opening your mouth opens the jaw, and holding up a hand shows its bones. The XDA report lists the main features:

• Skull, brain, muscles and eyes as separate layers, in an X-ray look or a solid look.
• Mixed-reality views: half real face and half interior, or an X-ray "lens" that follows your fingertip.
• Optional labels for bones, brain regions and muscles.
• Sliders to fit the skull to your head and soften the edges.

The live page has a Start camera button. The controls I could confirm in the demo cover layer selection, X-ray or solid rendering, brightness, the mixed-reality reveal, edge feathering, and the model's size, width, height, depth and position.

What "X-ray" does not mean​

The XDA report is plain about this: a regular store-bought webcam can't become a real-time X-ray machine. The tool takes a standard model and maps it to your body. You are seeing an average anatomy, not yours. The tool cannot show a fracture, an implant, a dental problem or any medical condition. Treat it as an interactive anatomy visual, not a diagnostic tool.

The sliders only change how the model is drawn. They are manual presentation adjustments, not anatomical measurement.

How it reportedly works​

The maker's account, which I could not read directly because the Reddit post was unavailable to me, is that this is a single HTML page. MediaPipe handles face and hand tracking in the browser, and three.js draws the 3D models. The maker also says Claude found the BodyParts3D anatomy dataset, simplified the models and aligned them to the face tracker's reference face. The first working version reportedly took under an hour. All of this comes from the maker and has not been independently measured.

On privacy, the maker says video is processed on your computer and never uploaded. A community list of Fable 5.2 demos repeats that claim with "the video never leaves your computer." That is the creator's statement, not an audit. If it matters to you, check your browser's network tab, or open the page in a profile with no sensitive tabs.

The maker credits the anatomy models to BodyParts3D, © The Database Center for Life Science, under CC BY-SA 2.1 Japan. That licence matters if you plan to reuse the assets. The evidence I have doesn't establish the licence for the rest of the code.

Trying it​

1. Open the public GitHub Pages demo. The maker says it is free and needs no account.
2. Click Start camera and allow camera access when your browser asks.
3. Choose a layer (skull, brain, muscles or everything) and X-ray or solid rendering.
4. Use the fit sliders to line the skull up with your head.
5. Try the mixed-reality split view or the fingertip lens, and raise a hand to see the hand bones.

Before you start, close other apps that use the camera. Use decent front lighting and keep your face fully in frame.

Known limits​

The maker invited feedback on how well the skull aligns across different faces, which acknowledges that fit varies. Commenters on a Reddit discussion summary reported imperfect alignment, trouble with jaw and teeth motion, odd hand orientation, and beards interfering with tracking. These are user reports, not formal tests. Expect it to work better for some faces and movements than others, and don't assume it behaves the same in every browser or on every device.

Why it matters​

The demo is a neat example of how little code a camera-based AR prototype now needs. Browser-side tracking, a WebGL renderer and an open anatomy dataset are enough. The harder part is knowing what the output means.

Even a polished overlay can look more authoritative than it is. Anyone building or sharing a tool like this should label it clearly as an educational visualization, which this one does.