| --- |
| title: X-Ray Anything |
| emoji: 🩻 |
| colorFrom: indigo |
| colorTo: blue |
| sdk: gradio |
| sdk_version: 6.17.3 |
| app_file: app.py |
| pinned: false |
| license: other |
| short_description: Photograph any object — watch it ride into a 3D scanner |
| tags: |
| - track:wood |
| - achievement:offbrand |
| - achievement:offgrid |
| models: |
| - black-forest-labs/FLUX.2-klein-9B |
| - Qwen/Qwen2.5-VL-3B-Instruct |
| --- |
| |
| # 🩻 X-Ray Anything |
|
|
| **Photograph any object. Watch it ride the patient table into a 3D MRI |
| machine. It comes back out as a radiograph of its insides.** |
|
|
| Drop in a photo of your keyboard, your coffee machine, your graphics card — |
| the scanner produces an X-ray view of *your actual object*: translucent |
| shell, glowing internals, silhouette preserved. |
|
|
| ## How the scan works |
|
|
| 1. **Qwen2.5-VL-3B** looks at your photo, identifies the object, and lists |
| the internal components it should contain |
| 2. **FLUX.2-klein-9B** *edits your photo* — same silhouette, same position — |
| into an X-ray radiograph revealing those internals (4-step rectified |
| flow, sub-second on the GPU) |
| 3. The whole experience plays out in a **3D scanner built with Three.js**: |
| your photo slides into the bore on the patient table, scan beams sweep, |
| the machine hums, and the radiograph slides back out |
|
|
| Everything fits the small-model constraint: 9B + 3B, both far under 32B. |
|
|
| ## Built on `gr.Server` |
|
|
| No default Gradio UI anywhere — the frontend is a single hand-written page |
| (Three.js scene, WebAudio scanner hum, boot sequence) served by **Gradio 6's |
| `gr.Server`**, with the model behind a `@app.api` endpoint and called via |
| `@gradio/client`. Queueing and concurrency still come from Gradio under the |
| hood. |
|
|
| ``` |
| frontend/index.html the 3D radiology console (no build step, CDN ESM) |
| app.py gr.Server: /xray endpoint + static routes |
| ``` |
|
|
| Run the UI locally without a GPU: `XRAY_MOCK=1 python app.py` replays a |
| cached scan so the full animation is testable on a laptop. |
|
|
| > 🎨 Radiographs are an AI's imagination of the inside, made for wonder — |
| > not engineering documentation. |
|
|
| *Build Small Hackathon 2026 · Thousand Token Wood track · |
| [Barath](https://huggingface.co/Barath)* |
|
|
| ## Submission |
|
|
| - **Demo video:** https://youtu.be/kVkX3Dg5gOg |
| - **Social post:** https://x.com/barathishere/status/2066882090054930704 |
|
|
| ### Badges claimed |
|
|
| | Badge | Why X-Ray Anything qualifies | |
| |---|---| |
| | 🌲 **Thousand Token Wood** (`track:wood`) | A whimsical, joyful toy: photograph anything, watch it ride into a 3D scanner and come back as a radiograph. | |
| | 🎨 **Off Brand** (`achievement:offbrand`) | No default Gradio UI — a hand-written Three.js radiology console (3D scanner, WebAudio hum, boot sequence) served via `gr.Server`. | |
| | 🔌 **Off the Grid** (`achievement:offgrid`) | Both models run in-Space on ZeroGPU; no external AI API is called. | |
|
|
