blender-python-stubs/README.md

blender-python-stubs

Strict type stubs for the Blender Python API. Provides autocomplete, type checking, and inline documentation for bpy, mathutils, bmesh, gpu, freestyle, and all other Blender Python modules. Generated stubs pass basedpyright all mode with 0 errors and use zero typing.Any.

Installation

pip install blender-python-stubs

The version matches your target Blender version:

Blender Version	Install Command	PyPI
5.1	pip install "blender-python-stubs>=5.1,<5.2"
5.0	pip install "blender-python-stubs>=5.0,<5.1"
4.5	pip install "blender-python-stubs>=4.5,<4.6"
4.4	pip install "blender-python-stubs>=4.4,<4.5"
4.3	pip install "blender-python-stubs>=4.3,<4.4"
4.2	pip install "blender-python-stubs>=4.2,<4.3"
4.1	pip install "blender-python-stubs>=4.1,<4.2"
4.0	pip install "blender-python-stubs>=4.0,<4.1"

Usage

Install alongside your Blender addon project for type checking with mypy, pyright, basedpyright, ty or your IDE.

import bpy

# Full autocomplete and type checking
obj = bpy.context.active_object
assert obj is not None
obj.location.x = 1.0

# Collection types with proper methods
bpy.data.objects.new("Cube", bpy.data.meshes.new("Mesh"))
bpy.data.images.remove(bpy.data.images["old"])

# GPU module fully typed
import gpu
shader = gpu.shader.from_builtin('UNIFORM_COLOR')
gpu.state.blend_set('ALPHA')

Features

• Full Blender API coverage — bpy.types, bpy.ops, bpy.props, bpy.utils, bpy.path, bpy.app, bpy.msgbus, mathutils, bmesh, gpu, gpu_extras, bpy_extras, freestyle, aud, blf, bl_math, imbuf, idprop
• Accurate collection types — bpy.data.objects returns BlendDataObjects (not generic bpy_prop_collection), exposing new(), remove(), and other collection-specific methods
• Readonly properties — uses @property decorators for readonly RNA attributes
• Context members — bpy.context.active_object, selected_objects, edit_object, and ~100 other screen context attributes are properly typed
• Constructor signatures — GPU types, mathutils types (Vector, Matrix, Euler, etc.) have typed __init__ methods
• Literal enum types — string parameters like gpu.state.blend_set(mode) use Literal["NONE", "ALPHA", ...] instead of plain str
• Docstrings — inline documentation on properties, methods, and functions
• Classmethod detection — Matrix.Identity(), Vector.Fill(), etc. correctly typed as @classmethod
• Dunder methods — __getitem__, __len__, __add__, __matmul__, and other operators introspected with correct return types (e.g. Matrix[0] returns Vector)
• Operator metadata — bpy.ops.mesh.primitive_cube_add.poll(), .idname(), .get_rna_type(), and .bl_options are typed via introspected _BPyOpsSubModOp wrapper
• GPU uniform types — shader.uniform_float() accepts Matrix, Vector, Euler, Quaternion, and Color in addition to float | Sequence[float], matching the C implementation
• Typed Context.copy() — returns a ContextDict TypedDict instead of dict[str, object], so unpacking into context.temp_override(**ctx) type-checks correctly
• Dynamic array types — Image.pixels typed as bpy_prop_array[float] (not list[float] or float)
• Zero Any usage — precise types throughout, no typing.Any fallbacks
• basedpyright all mode — 0 errors on generated stubs (4.0+), validated with typeCheckingMode: "all" (the strictest setting)

How It Works

Stubs are generated by running introspection inside Blender itself. A script runs in Blender's embedded Python interpreter using --background mode and collects:

1. RNA type definitions via rna_info.BuildRNAInfo() — all bpy.types classes, their properties, methods, inheritance, and readonly status
2. C extension signatures via inspect.signature() and RST docstring parsing — for mathutils, bmesh.types, gpu.types, aud, etc.
3. Screen context members via dir(bpy.context) — dynamically injected context attributes with type inference from runtime values, hardcoded overrides, and name-pattern heuristics
4. Collection wrapper classes via RNA srna attributes — maps bpy.data.objects to BlendDataObjects(bpy_prop_collection[Object]) instead of generic bpy_prop_collection[Object]

The introspection data is then passed through a stub generator that handles type cleaning, import resolution, docstring formatting, and black formatting.

Comparison with Alternatives

vs fake-bpy-module

Feature	blender-python-stubs	fake-bpy-module
@property for readonly	Yes (1700+)	No
Any usage	0	1800+
Collection wrapper types	BlendDataObjects	bpy_prop_collection[Object]
bpy_struct methods	Introspected	Missing
rna_type attribute	Yes	Missing
Operator metadata	poll(), idname(), etc.	Missing
Constructor __init__	Yes (mathutils, gpu, etc.)	No
Literal enum types	Yes	Yes (via stub_internal)
Context members	~100 typed	~100 typed
basedpyright all mode	0 errors (4.0+)	Not tested
Docstrings	Yes	Yes

vs bpystubgen

bpystubgen generates stubs from Blender's Sphinx documentation, not from runtime introspection. This means it can miss C-level methods, dynamic context members, and runtime-only type information. Our approach introspects the actual running Blender instance, ensuring stubs match the real API.

Supported Versions

Blender 4.0 through 5.1. Each Blender version gets its own stub package version.

Generating Stubs

To generate stubs for a specific Blender version, run:

uv run poe generate 5.1

This will automatically download the corresponding Blender executable and run the introspection and stub generation process. You can also type-check the generated stubs:

uv run poe typecheck-stubs 5.1

Contributing

Pull requests are welcome. The project uses:

• uv for dependency management
• poe for task running
• basedpyright for type checking
• black for formatting
• ruff for linting

# Run all checks (format, lint, typecheck, tests)
uv run poe check

Disclaimer

This project was coded with assistance of AI.

License

MIT

---

Made with ❤️ at Autour de Minuit (ADV)