Skip to content

kws/invariant-gfx

Repository files navigation

Invariant GFX

A deterministic, functional graphics pipeline built on Invariant. Invariant GFX allows developers to build complex visual assets (icons, badges, dynamic UI components, Stream Deck buttons, data visualizations) by plugging together reusable "pipeline parts" in a DAG-based system.

Note: This project builds on Invariant, a deterministic execution engine for DAGs. For information about Invariant's core concepts (DAG execution, caching, execution model, parameter markers, etc.), see the upstream README and Invariant documentation.

What is Invariant GFX?

Invariant GFX gives you aggressive caching and deduplication: identical visual operations run once and are reused. Outputs are bit-for-bit reproducible, and layout can be driven by upstream artifact dimensions so the same graph scales across sizes.

High-level features:

  • Smart layout — Ops inspect upstream artifact dimensions to calculate positions dynamically.
  • Anchor-based composition — Position layers with absolute() and relative() builder functions, referencing previously-placed named layers.
  • Content-sized layout — Row/column flow with automatic sizing.
  • Effects as subgraphs — Filter primitives (blur, alpha, morphology) are composed into effect recipes (e.g. drop shadow) via SubGraphNode. See docs/effects.md.
  • Deterministic and cacheable — No floats in cacheable data; all layout uses Decimal or integers.

Relationship to Invariant: Invariant GFX is a child project of Invariant. The parent provides the DAG execution engine, caching, and core protocols (no image awareness). Invariant GFX provides graphics ops (gfx:render_text, gfx:composite, gfx:render_svg, etc.) and artifacts (ImageArtifact, BlobArtifact). It uses Invariant's Executor and store directly—no wrapper.

Get started

Installation

pip install invariant-gfx

To run the bundled examples or any code that hardcodes a font family (such as the Inter snippet below), install the optional font pack — it pulls justmytype-western-core, which registers Inter, Noto Sans, JetBrains Mono, and a few more with JustMyType:

pip install 'invariant-gfx[fonts]'

For development inside this repository:

git clone https://github.com/kws/invariant-gfx
cd invariant-gfx
uv sync --all-groups

The dev group also installs justmytype-western-core, so the test suite and examples find a font without extra setup.

Quick Start

Minimal example: text on a solid background, with proportional font sizing (14pt at 72px reference). The snippet below uses Inter, which ships via the fonts extra; substitute any font installed on your system.

from decimal import Decimal
from invariant import Executor, Node, ref
from invariant.registry import OpRegistry
from invariant.store.memory import MemoryStore

from invariant_gfx import register_core_ops
from invariant_gfx.anchors import relative

registry = OpRegistry()
register_core_ops(registry)
store = MemoryStore()
executor = Executor(registry=registry, store=store)

size = 72
font_size = int(Decimal(str(size)) * Decimal("14") / Decimal("72"))

graph = {
    "text": Node(
        op_name="gfx:render_text",
        params={
            "text": "Hello",
            "font": "Inter",
            "size": font_size,
            "color": (255, 255, 255, 255),
        },
        deps=[],
    ),
    "background": Node(
        op_name="gfx:create_solid",
        params={
            "size": (size, size),
            "color": (40, 40, 40, 255),
        },
        deps=[],
    ),
    "final": Node(
        op_name="gfx:composite",
        params={
            "layers": [
                {"image": ref("background"), "id": "background"},
                {"image": ref("text"), "anchor": relative("background", "c@c"), "id": "text"},
            ],
        },
        deps=["background", "text"],
    ),
}

results = executor.execute(graph, ["final"])
results["final"].image.save("output.png", format="PNG")

Proportional sizing can also be driven by context + CEL when using the template+context pattern; see docs/architecture.md.

Examples

  • examples/quick_start.py — Minimal text-on-background with proportional sizing.
  • examples/thermometer_button.py — Icon + text + layout + composite pipeline.
  • examples/text_badge.py — Dynamic SVG resizing driven by text dimensions.
  • examples/color_dashboard.py — Multi-cell dashboard with nested layouts.
  • examples/text_drop_shadow.py — Text with drop-shadow recipe (effect subgraph).
  • examples/effects_showcase.py — All effect recipes (drop shadow, outer stroke, outer glow, inner shadow, inner glow, reflection) in a grid.
  • examples/shapes_showcase.py — Shapes library (rect, circle, etc.) with render_svg.
  • examples/serialized/square_button_badge.json — Parameterized serialized graph document for a fit-width text badge.

Where to go next: See docs/README.md for the documentation index. For op specs and the template+context pattern, see docs/architecture.md. For reference pipelines, see docs/reference_pipelines.md. For effects and filter primitives, see docs/effects.md.

Reference

Invariant GFX provides graphics ops under the gfx: namespace: sources (resolve_resource, create_solid), transformers (render_svg, render_text, resize), composition (composite, layout), casting (blob_to_image), and effects (extract_alpha, blur, colorize, translate, pad, etc.). See docs/architecture.md and docs/effects.md for the full list and specifications.

Contributing

# Run tests
uv run pytest

# Lint
uv run ruff check src/ tests/

# Format
uv run ruff format src/ tests/

For constraints, terminology, and implementation context, see AGENTS.md.

For execution model and GFX-specific design, see docs/architecture.md.

License

MIT License — see LICENSE for details.

About

Invariant GFX is a deterministic, functional graphics pipeline built on Invariant for creating composable visual assets such as icons, badges, dashboards, and UI components. It combines DAG-based composition, smart layout, reusable effect subgraphs, and caching to produce reproducible renders.

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages