docs: make the README a product page, with generated screenshots

README.md

@@ -1,10 +1,81 @@
 # MatteLoop
 
-MatteLoop is a desktop app for previewing local videos, removing their
-backgrounds, and exporting lossless transparent animated WebP files. V1
-targets macOS 15+ arm64 and Windows x64; Linux artifacts remain deferred. The
-current native qualification covers macOS 15+ arm64 only, and native artifacts
-are unsigned.
+MatteLoop cuts the background out of a video on your own machine and gives you
+back a transparent animated WebP that loops.
+
+![The MatteLoop main window: a source video on the left with a crop rectangle,
+the cut-out result on the right over a transparency checkerboard, and the
+inspector on the far right](assets/screenshots/main-window.png)
+
+Nothing leaves your computer. The video, the model weights, every intermediate
+frame and the finished loop stay on your disk — there is no account, no upload,
+and no network call except the one that fetches a model weight the first time
+you use it.
+
+## What it is for
+
+You have a clip of a subject — a person, a pet, a product — and you want it
+without its background, as a loop you can drop onto any page or into any
+composite. MatteLoop is the whole path from that clip to that loop:
+
+- **Open a local video** and scrub to the part you want.
+- **Pick the frames** with IN and OUT points on the timeline.
+- **Preview a single frame** to judge the cutout before committing to a render.
+- **Trim, crop and resize** the finished cut, and watch it loop on screen until
+  it is right.
+- **Render** a lossless transparent animated WebP.
+
+The result is lossless and genuinely transparent — an alpha channel, not a
+matte painted onto a colour.
+
+### Judge the loop before you render it
+
+The transform stage works on the cut you already have. Trimming, cropping and
+resizing never touch the stored frames; they are a specification applied when
+the file is encoded, so you can change your mind as often as you like without
+segmenting anything again.
+
+![The inspector's Transform section expanded, showing trim first and last
+frame, crop position and size, and resize width, height and
+percent](assets/screenshots/transform-expanded.png)
+
+### Choose the model that suits the shot
+
+Different subjects want different weights. MatteLoop ships a catalog of them
+and downloads only what you ask for, so a first run costs one model rather than
+the whole set. The manager shows what is on disk, what a weight costs, and
+which weights are left over from an earlier version of the segmentation
+runtime.
+
+![The model manager listing the model catalog with size, cache status and which
+model is active, and buttons to download, remove or re-download outdated
+weights](assets/screenshots/model-manager.png)
+
+Weights are not bundled with the application. They download on first use and
+are cached locally; `birefnet-portrait` alone is about 928 MiB and the complete
+catalog is about 6.35 GiB, which is exactly why you fetch them one at a time.
+
+## Built on rembg
+
+The segmentation in MatteLoop is [rembg](https://github.com/danielgatis/rembg)
+by Daniel Gatis. The model catalog, the session classes that run each
+architecture, and the whole approach to background removal come from that
+project, and the weights MatteLoop downloads are its release artifacts, listed
+with their checksums in `resources/model-manifest.json`. MatteLoop is the
+desktop application around it: the timeline, the preview, the transform stage
+and the WebP encoder.
+
+The authors of the individual model architectures behind those weights are
+credited in [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md), along with every
+other component and its licence.
+
+## Getting it
+
+Native builds for macOS 15+ arm64 and Windows x64 are published on the
+[releases page](https://github.com/smb/MatteLoop/releases). They are unsigned:
+Windows warns through SmartScreen, and macOS needs a right-click Open the first
+time. Linux artifacts are deferred, and the current native qualification covers
+macOS 15+ arm64 only.
 
 ## Run from source
 
@@ -15,14 +86,10 @@ uv sync --frozen --all-groups
 uv run matteloop
 ```
 
-The 13 V1 models come from the upstream rembg release artifacts listed in
-`resources/model-manifest.json`. They download on first use, are cached
-locally, and are not bundled: `birefnet-portrait` is about 928 MiB and the
-complete V1 catalog is about 6.35 GiB.
-
 For quick diagnostics, `uv run matteloop --version` avoids Qt and model access;
 `uv run matteloop --smoke-test` exercises the local runtime without downloading
-weights.
+weights; and `uv run matteloop --providers` reports which ONNX Runtime
+execution providers the installed runtime actually offers.
 
 Run the checks with:
 
@@ -35,6 +102,17 @@ QT_QPA_PLATFORM=offscreen uv run pytest -q
 Model weights and generated workspaces are intentionally local and are never
 committed.
 
+The screenshots above are generated rather than collected, so a layout change
+never leaves them quietly out of date. Regenerate them with:
+
+```sh
+QT_QPA_PLATFORM=offscreen uv run python scripts/screenshots.py
+```
+
+It decodes `assets/demo/golden-retriever.mp4`, segments it with
+`birefnet-general-lite` — which has to be in the local model cache already —
+and rewrites `assets/screenshots/`.
+
 New installs use the MatteLoop cache and workspace names. Existing model
 weights, compiled provider caches and thumbnails under
 `~/Library/Caches/rembggui/` and promoted cuts in an output directory's
@@ -50,6 +128,9 @@ instructions. See [docs/building.md](docs/building.md) for local and manual
 Actions commands, the five inseparable distribution outputs, unsigned-artifact
 launch warnings, and current platform qualification status.
 
+Notes for people writing decoder tests live in
+[docs/testing-fixtures.md](docs/testing-fixtures.md).
+
 ## License
 
 MatteLoop's original source code, documentation, and visual assets are licensed
@@ -68,12 +149,3 @@ and [LGPL version 3](legal/LGPL-3.0.txt) texts, prominent
 [Qt/PySide notice](legal/QT-PYSIDE-LGPL-NOTICE.md), and practical
 [replacement instructions](legal/RELINK.md) are kept in the repository and
 inside every native app; they do not change MatteLoop's 0BSD license.
-
-### Synthetic fixture rotation
-
-`tests.fixtures.media_factory.make_video(..., rotation=...)` writes a sorted,
-versioned adjacent sidecar named `<video>.matteloop.json`. It records the
-counter-clockwise presentation rotation as `rotation_ccw`. The locked PyAV 16
-wheel cannot author portable MP4 display-matrix metadata, so future synthetic
-source-decoder tests must consume this explicit fixture contract instead of
-expecting a display matrix in the video container.

docs/testing-fixtures.md

@@ -0,0 +1,10 @@
+# Test fixtures
+
+## Synthetic fixture rotation
+
+`tests.fixtures.media_factory.make_video(..., rotation=...)` writes a sorted,
+versioned adjacent sidecar named `<video>.matteloop.json`. It records the
+counter-clockwise presentation rotation as `rotation_ccw`. The locked PyAV 16
+wheel cannot author portable MP4 display-matrix metadata, so future synthetic
+source-decoder tests must consume this explicit fixture contract instead of
+expecting a display matrix in the video container.