Yerner's asset generator — turn any photo into engraved goth cutouts, dreamy film scans, and rotating obsidian sculptures. Entirely on-device, no internet needed.
Assets: of yearners is a fully offline goth-asset generator. Pick a photo, choose a mode, and the app produces a stylized asset you can save straight to your gallery. There are no accounts, no uploads, and no network calls — the ML model and every filter run on your device.
| # | Mode | Output |
|---|---|---|
| 01 | Goth Cutout | Engraved, grainy B&W die-cut of the subject on a transparent square canvas — background removed, autocontrast, heavy grain |
| 02 | Mystic Film | Dreamy analog B&W photo — square crop, lifted hazy tone, soft glow, film grain, dark border |
| 03 | Background Removal | Just the subject, cut out with a soft alpha mask — transparent PNG |
| 04 | 3D Sculpt BETA |
The subject's silhouette reimagined as a glossy black rotating sculpture, exported as a looping GIF |
Everything happens in three files of Dart — no backend, no SDKs:
- Decode — the picked image is decoded and EXIF-rotated in a background isolate.
- Segment — a bundled u2netp ONNX model (~4.5 MB)
runs on-device via
onnxruntimeto produce a soft alpha mask of the subject. Transparent inputs are flattened onto white first so dark subjects survive segmentation. - Stylize — pure-Dart pixel pipelines (
package:image) apply the chosen mode:- Goth Cutout / Background Removal use the mask directly (hard die-cut vs. soft alpha).
- Mystic Film skips the mask and works on tone alone.
- 3D Sculpt converts the silhouette into a signed distance field (Felzenszwalb–Huttenlocher distance transform), inflates it with a circular profile, bends it in depth with random sine warps, and renders 24 turntable frames with a tiny raymarcher — then encodes them as a looping GIF.
Every heavy step runs in a Dart isolate, so the UI stays smooth and the progress
label ticks through each stage (removing background, engraving, frame 12/24, …).
The interface follows an editorial monochrome design system (see design.md):
pure black/white with felt-gray accents, sharp corners everywhere except full-pill buttons
and tags, no shadows, whisper-weight display type, and a single chromatic gesture — the
animated iridescent hero (sage → amber → oxblood) that collapses into the navbar on scroll.
Android — grab the latest APK from Releases and sideload it. Release builds are produced by the GitHub Actions workflow and signed.
Linux desktop — build from source (below).
Requires Flutter (stable channel, Dart SDK ≥ 3.12).
git clone https://github.com/Project-Delilah/assets_of_a_yearner-flutter
cd assets_of_a_yearner-flutter
flutter pub get
flutter run -d linux # desktop
flutter build apk # Android (needs your own signing config, see below)Release signing reads android/key.properties; CI recreates that file from repository
secrets (STORE_PASSWORD, KEY_PASSWORD). For a local release build, point it at your
own keystore.
Launcher icons are generated from assets/icons/ with:
dart run flutter_launcher_iconsflutter test # unit tests (filters, sculpt, isolates)
flutter test integration_test/onnx_pipeline_test.dart -d linux # real ONNX inferenceThe integration test runs the actual u2netp model on fixture photos and writes its masks
and rendered sculpt frames to build/it_out/ for eyeballing.
lib/
main.dart # all UI + isolate trampolines for the processing pipeline
filters.dart # BgRemover (ONNX u2netp) + gothify / mystify / cutout filters
sculpt3d.dart # silhouette → SDF → raymarched turntable GIF
assets/
models/u2netp.onnx # bundled segmentation model
examples/ # mode preview images
icons/ # launcher icon set
design.md # the design system the UI follows
Sapan Gajjar — github.com/isg32
GPL-3.0 — free as in freedom. If you ship a modified version, ship the source too.