Streamlit Drawable Konva

Streamlit custom component (Components v2) that provides a sketching canvas using Konva.js / react-konva.

The public Python API is intentionally compatible with streamlit-drawable-canvas-fix (st_canvas → CanvasResult(image_data, json_data)). Object JSON uses Konva-oriented fields (x/y/points/…), not Fabric.js schemas.

Features

• Freehand, line, rect, circle, point, polygon drawing
• Transform mode (move / scale / rotate); double-click to delete
• Background color or image
• Realtime or on-demand updates to Streamlit
• Undo / redo / clear toolbar
• Returns RGBA image_data and scene json_data

Why Konva.js (vs Fabric.js)?

Konva is not universally “better” than Fabric — both are mature 2D canvas libraries. For this Streamlit component, Konva is the better fit.

Why we chose Konva here

1. React + Streamlit Components v2 — the official CCv2 React template maps cleanly onto react-konva. Drawing state stays in React; updates go out via setStateValue without fighting an imperative canvas singleton.
2. Layered export — background images can live on a non-exported layer while strokes/shapes export cleanly to image_data, matching the upstream “background is not round-tripped” behavior with less glue code.
3. Maintainability — freerdraw / shapes / polygon / transform are ordinary React event handlers and Konva nodes, which is easier to evolve next to a modern Vite + TypeScript frontend than a Fabric-centric CRA-era stack.
4. Performance for annotation UIs — Konva’s layer/stage model scales well for ROI / sketching workloads typical in Streamlit apps.

When Fabric may still win

• You need byte-compatible Fabric JSON (existing pipelines, demos, or tutorials that assume left/top/path Fabric schemas).
• You want Fabric’s mature free-transform / skew / complex text editing out of the box without reimplementing edge cases.

This package keeps a compatible Python API (st_canvas → CanvasResult) so apps can migrate from streamlit-drawable-canvas-fix, while accepting that json_data is Konva-oriented rather than a Fabric dump.

Install (uv)

uv sync

Build the frontend once (required before first Streamlit run):

cd streamlit_drawable_konva/frontend
npm install
npm run build
cd ../../

Editable install is handled by uv sync via the local pyproject.toml.

Demo

uv run streamlit run app.py

Development

CCv2 watches a production-style build into frontend/build (no separate webpack dev server URL). Use two terminals:

Terminal A — frontend watch build

cd streamlit_drawable_konva/frontend
npm install
npm run dev

Terminal B — Streamlit

uv sync
uv run streamlit run app.py

Refresh the browser after frontend rebuilds.

API

from streamlit_drawable_konva import st_canvas

result = st_canvas(
    fill_color="rgba(255, 165, 0, 0.3)",
    stroke_width=3,
    stroke_color="#000000",
    background_color="#eee",
    background_image=None,
    update_streamlit=True,
    height=400,
    width=600,
    drawing_mode="freedraw",  # freedraw|line|rect|circle|point|polygon|transform
    initial_drawing=None,
    display_toolbar=True,
    point_display_radius=3,
    key="canvas",
)

# result.image_data -> np.ndarray | None
# result.json_data  -> dict | None

Polygon controls

• Left-click: add point
• Right-click: close polygon
• Double-click: remove latest point

Transform controls

• Click object to select
• Drag / resize / rotate with the transformer
• Double-click selected object to delete

Packaging notes

• Registered as streamlit-drawable-konva.st_canvas via the in-package streamlit_drawable_konva/pyproject.toml manifest (asset_dir = frontend/build).
• Requires Streamlit >= 1.51 (Components v2).
• Frontend toolchain targets Node >= 18 (Vite 5).

Publishing (GitHub / PyPI / Streamlit Cloud / gallery)

Step-by-step instructions: PUBLISHING.md.

Order: push to GitHub → publish to PyPI → deploy app.py on Community Cloud → submit gallery/streamlit-drawable-konva.json as a PR to streamlit/gallery.

License

MIT