Unbagrnd v0.5.6

A patch release fixing two errors that appeared when processing images on Android (batch mode) and on desktop (large models).

Bug fixes

• Android: batch processing failed with could not access content://... No such file or directory.
  Photos chosen through the system photo picker now process correctly in Batch mode. Previously the app tried to stat the content:// URI as if it were a regular file path, which always failed.
• Desktop: Failed to allocate memory for requested buffer of size ... on large models.
  1024px models (notably BiRefNet) used to request one very large memory block up front and failed wheory is now allocated on
  demand and released immk usage considerably.

Improvements

• When memory really is i suggests closing other
  apps or switching to a eta, instead of showing the
  raw ONNX Runtime messag
• The model session is refailure, so the next image
  in a batch doesn't fail

Internal

• Version bumped to 0.5.6 core, API).
• Added a unit test covering content:// URI handling.

Notes

The memory fix passes buiot yet been verified with a real BiRefNet inference rf a similar error stillshows up, please report i image size.

What's new in v0.5.5

• Toasts confirm saves, ZIP exports, batch results and failures, with a "Show" shortcut to the file.
• Fix: the Refine editor no longer gets stuck after Clear, and Erase/Restore switch freely while strokes are pending.

Also in 0.5.x: alpha-mask export, edge refinement (shift/smooth/feather) and batch upgrades (shared background/shadow, file-name templates, cancel/retry, ZIP export).

What's new in v0.5.0

• Export mask - save the cutout's alpha as a black-and-white matte for layer masks.
• Edge refinement - shift, smooth and feather the outline with a live preview (great for hair and fur).
• Batch upgrades - shared background/shadow, file-name templates, cancel, retry failed, and ZIP export.
• Android - layouts tuned for small screens.

The REST API and model cache are unchanged.

feat: add self-hosted REST API with batch endpoint, restructure into Cargo w
orkspace (v0.4.0)

- Split shared background-removal logic into a new unbagrnd-core crate
  (model catalog, download/caching, ONNX inference), reused as-is by
  both the Tauri desktop app and the new API - no duplicated logic.
- Add unbagrnd-api: a self-hosted Axum REST API with API-key auth
  (SQLite-backed, SHA-256 hashed, never stored plaintext), admin-key
  gated key management, per-key rate limiting, bounded inference
  concurrency, upload size/dimension limits, and OpenAPI/Swagger docs
  at /docs.
- Add POST /v1/remove-background/batch: up to UNBAGRND_MAX_BATCH_SIZE
  images per request, each processed independently so one bad file
  doesn't fail the rest - results come back as a JSON array in upload
  order, still bounded by the same inference concurrency semaphore.
- Add Dockerfile (Debian trixie base - bookworm's libstdc++ is too old
  for the prebuilt ONNX Runtime ort downloads), docker-compose.yml,
  .env.example, and a Postman collection for one-command self-hosting.
- Update README with full API documentation.
- Bump version to 0.4.0 across the workspace.
- Desktop and Android builds verified unaffected by the restructure.

See the assets below to download and install unbagrnd for your platform.

unbagrnd removes image backgrounds entirely on-device: no cloud API,
no account, and no telemetry. Pick from several background-removal
models in Settings; each one downloads once on first use (starting
at ~43 MB for the default) and is cached locally after that.

See the assets below to download and install unbagrnd for your platform.

unbagrnd removes image backgrounds entirely on-device: no cloud API,
no account, and no telemetry. Pick from several background-removal
models in Settings; each one downloads once on first use (starting
at ~43 MB for the default) and is cached locally after that.

See the assets below to download and install unbagrnd for your platform.

unbagrnd removes image backgrounds entirely on-device: no cloud API,
no account, and no telemetry. Pick from several background-removal
models in Settings; each one downloads once on first use (starting
at ~43 MB for the default) and is cached locally after that.

See the assets below to download and install unbagrnd for your platform.

unbagrnd removes image backgrounds entirely on-device: no cloud API,
no account, and no telemetry. Pick from several background-removal
models in Settings; each one downloads once on first use (starting
at ~43 MB for the default) and is cached locally after that.