Krita Smart Select

Krita Smart Select is a Linux plugin for Krita 5.3.1 that creates soft AI-assisted selections from a user-drawn region.

It is designed as a local Krita AI selection, object selection and alpha matting plugin for digital artists.

The tool works like a native lasso selection tool first: draw or adjust the area around an object, then press Select object with AI. The worker runs BiRefNet HR Matting on the selected crop and applies the returned soft alpha as a Krita selection.

Features

• Native Krita canvas tool and Tool Options integration.
• Replace/add/subtract/intersect selection actions through Krita selection semantics.
• Soft alpha selection masks instead of binary-only selections.
• Managed user-level Python runtime for inference dependencies.
• Pinned BiRefNet HR Matting model snapshot cache.
• Local structured logs with OpenTelemetry-compatible field names.
• Debug image export for local diagnosis when explicitly enabled.

Examples

Before	After

Smart Select preserves transparent surfaces and subtle see-through details, such as eyeglass lenses.

Before	After

Soft alpha selection keeps fine hair detail instead of reducing the subject to a hard binary cutout.

Before	After

Before	After

Requirements

For users:

• Linux.
• Krita 5.3.1.
• uv available in PATH.
• Network access for the first runtime and model download.
• NVIDIA driver with CUDA support for GPU inference, or CPU fallback when CUDA is unavailable.
• Enough free disk space for the managed runtime and model cache.

For development:

• CMake.
• C++20 compiler.
• Krita 5.3.1 source tree and build tree.
• Qt5/KF5/Krita development dependencies required by that Krita build.
• Python with uv.
• clang-format.

The plugin does not install PyTorch, model dependencies or model checkpoints inside the plugin source tree. Runtime data is stored under Krita user data:

<krita-app-data>/smart_select/
  models/
  runtime/

User Installation

Download the Linux release ZIP:

krita-smart-select-linux-x64-<version>.zip

Install it from Krita:

1. Open Tools -> Scripts -> Import Python Plugin from File....
2. Select the release ZIP.
3. Restart Krita.
4. Enable Smart Select in the Python Plugin Manager if needed.
5. Select Smart Select from the toolbox.

The ZIP root contains the Python plugin descriptor and package:

smart_select.desktop
smart_select/
  SmartSelectTool.action
  lib/kritasmartselect.so
  worker/
  ...

On first use, draw or edit a region, press Select object with AI, and let the worker create the runtime and download the model.

The release also includes a .tar.gz archive with the expanded install layout for manual inspection or manual profile installation.

Development Setup

Create the development Python environment:

uv venv .venv
uv pip install --python .venv/bin/python -r requirements-dev.txt

Build against a Krita 5.3.1 source/build tree:

KRITA_SRC=/home/bogdan/Work/sources/krita \
KRITA_BUILD=/home/bogdan/Work/build/krita-5.3.1 \
BUILD_JOBS=20 \
./scripts/build-krita-5.3.1.sh

The build output is written to:

dist/krita-smart-select/

Install the local build into Krita:

mkdir -p ~/.local/share/krita/pykrita ~/.local/share/krita/actions
rm -rf ~/.local/share/krita/pykrita/smart_select
cp -a dist/krita-smart-select/pykrita/smart_select ~/.local/share/krita/pykrita/
cp dist/krita-smart-select/pykrita/smart_select.desktop ~/.local/share/krita/pykrita/
cp dist/krita-smart-select/share/krita/actions/SmartSelectTool.action ~/.local/share/krita/actions/

Package a Linux archive:

./scripts/package-linux.sh 0.1.0

This writes two files into dist/:

krita-smart-select-linux-x64-0.1.0.zip
krita-smart-select-linux-x64-0.1.0.tar.gz

Publish those archives to GitHub Releases:

./scripts/create-github-release.sh 0.1.0

The release helper expects a clean worktree, creates and pushes tag v0.1.0, then creates or updates the matching GitHub Release with both assets attached. This is a local release flow because the package contains a native kritasmartselect.so built against Krita 5.3.1. A fully hosted GitHub Actions release would need a matching Krita build environment or a self-hosted runner.

Verification

Run the default checks:

PYTHON=.venv/bin/python ./scripts/test-code.sh

This runs Python compile checks, Ruff, mypy, worker unit tests and the C++ core tests when the Krita source/build tree is available.

Run the opt-in real inference smoke test only when model/runtime setup is acceptable:

SMART_SELECT_RUN_BIREFNET_SMOKE=1 PYTHON=.venv/bin/python ./scripts/test-birefnet-smoke.sh

Documentation

• Product Scope
• Architecture
• Worker Protocol
• Snapshot And Temporary Files
• Models
• Runtime Environment
• Observability
• Typing
• Testing
• Code Style
• Licensing
• Contributing

Acknowledgements

Krita Smart Select builds on ideas, APIs and open-source work from:

• Krita and the KDE Krita project, whose native tool, selection and plugin APIs make this kind of canvas-integrated workflow possible.
• BiRefNet, especially the ZhengPeng7/BiRefNet_HR-matting model used for direct soft-alpha extraction.
• Krita Vision Tools, which provided a useful reference for ML-assisted Krita tooling and user-facing selection workflows.
• CorridorKey, which helped inform the product direction toward alpha-first, plug-and-play matting.

License

Krita Smart Select is licensed under GPL-3.0-only. See LICENSE and docs/licensing.md.