# tools/ — Offline asset pipeline

Offline tooling for preparing 3D assets served by the hmwebviewer. Nothing here
ships to the browser at runtime; these produce the optimized assets and
pre-rendered placeholders that the viewer consumes.

## preprocess.mjs — Draco + KTX2 compression

Compresses a `.glb`/`.gltf` into a single self-contained `.optimized.glb` using
[gltf-transform](https://gltf-transform.dev/): Draco for geometry, KTX2/Basis
Universal for textures, in one pass. Output loads directly in the viewer's
`GLTFLoader` (with `DRACOLoader` + `KTX2Loader` wired on the shared loader).

### Install (once)

These dev dependencies are **not** in `package.json` yet. Run before first use:

```bash
npm install -D @gltf-transform/core @gltf-transform/functions \
                @gltf-transform/extensions @gltf-transform/cli
```

The CLI package provides the KTX2 encoder wiring. gltf-transform fetches the
platform basis encoder automatically on first KTX2 run (network needed once).

### Usage

```bash
node tools/preprocess.mjs <input.glb> [output.glb] [--draco-bits N] [--ktx2|--no-ktx2]
```

- Default output: `<input>.optimized.glb` (e.g. `model.glb` → `model.optimized.glb`).
- `--draco-bits N` — position quantization bits, 8..16 (default **14**). Lower =
  smaller file, lossier geometry. Tune per asset.
- `--ktx2` / `--no-ktx2` — texture encoding toggle (default **on**).

Prints before/after byte sizes and reduction %.

### Outputs

- Compressed GLBs land in [`samples/`](../samples/) (`samples/*.optimized.glb`).
- Pre-rendered WebP placeholders land in [`public/previews/`](../public/previews/)
  (`*.webp`) — see the pre-render section below.

If a required package is missing at runtime, the script prints the install
command above and exits non-zero (no silent skip).

## Sample assets

Place source `.glb` files under `samples/` (suggested split: small / medium /
large). Compress with `preprocess.mjs` and commit the `.optimized.glb` outputs.
Samples are not committed yet — to be added when a sample asset is available.

## Pre-render pipeline (360° animated WebP placeholder)

For Path A (server asset) the server ships a pre-rendered 360° turntable so the
page shows motion instantly while the real model decodes in the background.
Output: `public/previews/<asset>.webp` (animated, with alpha).

**Status: to be implemented when a sample asset is available.** Two options:

1. **Blender headless CLI (preferred).** Camera parented to an empty at the
   model origin, Z axis 0→360° keyframed over N frames. Render frames:
   ```bash
   blender -b scene.blend -o //frame_### -f 1..N -F PNG
   ```
   Assemble to animated WebP (alpha) or WebM VP9 via ffmpeg:
   ```bash
   ffmpeg -framerate 24 -i frame_%03d.png -loop 0 -plays 0 out.webp
   ```

2. **Puppeteer + headless three.js (fallback).** Spin a minimal three.js page,
   rotate the model, and call `page.screenshot({ type: 'webp' })` per rotation
   step. Stitch frames into an animated WebP.

Both options need a committed sample asset first; blocked on P4-2.