Cinefractal

Cinefractal is a Python library for generating static and animated escape-time fractals. The number crunching is written in Rust (via PyO3 and maturin) and parallelised with rayon, while the public API is plain Python that returns NumPy arrays ready to hand off to Pillow or MoviePy.

The complete documentation can be found on github pages.

The library has three main building blocks:

1. Iteration field interface — generates an iteration field or a colorized image of a fractal from a handful of parameters.
2. Animation system — a key-frame animation system that drives the iteration field interface over time, ready to render to MP4 with MoviePy.
3. Focal (autofocus) system — searches for interesting parts of a fractal using a variance-based heuristic; the points it finds can steer the animation.

Gallery

Mandelbrot	Julia	Burning Ship	Tricorn	Celtic

Installation

pip install cinefractal

Cinefractal requires Python 3.13+ and pulls in NumPy automatically. The bundled examples additionally use, depending on what they do:

pip install pillow        # save/show images
pip install plotly pandas # visualize a raw iteration field
pip install moviepy       # render MP4 movies

Quick start

Render the classic Mandelbrot image to a file:

from PIL import Image
from cinefractal import IterationFieldInterface, ColorSystem

ifi = IterationFieldInterface()
ifi.set_numpy_extension(1024, 1280)        # rows (height), columns (width)
ifi.set_maximum_iterations(1000)
ifi.set_center_point(-1.0, 0.0)
ifi.set_extension(1.2)                      # smaller extension = deeper zoom
ifi.set_colorization_information(ColorSystem.cyclical(3.0), False)
ifi.set_colorization_log_strength(0.5)

color_field = ifi.get_color_field()         # numpy (rows, cols, 3) uint8
Image.fromarray(color_field).save("mandelbrot.png")

The three building blocks

Iteration field interface

IterationFieldInterface is the entry point for single images. It supports five escape-time fractals:

All of them take an integer exponent (>= 2) used in the iteration formula.

The field is configured through small setter methods, each with a sensible default:

You can then query the result in three forms:

• get_discrete_iteration_field() → 2-D uint16 array (iterations until escape)
• get_continuous_iteration_field() → 2-D float32 array (smoothed)
• get_color_field() → 3-D uint8 array (rows, cols, 3), ready for Pillow

Colorization offers several base color schemes — turbo, viridis, magma, plasma, inferno, cool, and cyclical(repetition_factor) — plus an optional logarithmic strength that boosts the resolution of low iteration counts:

from cinefractal import IterationFieldInterface, ColorSystem

ifi = IterationFieldInterface()
ifi.set_colorization_information(ColorSystem.viridis(), False)  # False = continuous
ifi.set_colorization_log_strength(0.5)

Animation system

The animation system is key-frame based and drives an IterationFieldInterface over time. You build the animation with an AnimationRecorder, freeze it into an immutable AnimationPlayer, then ask the player to apply the animated state for a given time before reading the color field.

from cinefractal import AnimationRecorder, IterationFieldInterface, FractalType, ColorSystem

anim = AnimationRecorder(exponent=2, log_strength=2.0, max_iterations=1000)
anim.set_keyframe_fractal_seed_value(0.0, -0.8, 0.156)  # constant Julia seed
anim.set_keyframe_extension(6.0, 1e-4)                  # zoom in over 6 s
player = anim.get_animation_player()

ifi = IterationFieldInterface()
ifi.set_fractal_type(FractalType.julia())
ifi.set_colorization_information(ColorSystem.inferno(), False)

player.apply_animation(3.0, ifi)        # state at t = 3 s
frame = ifi.get_color_field()           # render that frame

Key frames exist for the exponent, Julia seed, render center point, extension (zoom, interpolated in logarithmic space), maximum iterations, and color log strength. Combined with MoviePy's VideoClip, this renders straight to MP4 — see julia_zoom_movie.py.

Focal (autofocus) system

The focal system scatters candidate points across a search region and keeps the most interesting ones, judged by the local variance of the iteration count. FocalSystemInterface runs the (parallel) search and returns an immutable FocalPointResult, which can hand the points back either unordered (with their scores) or as an ordered path ready to drive an animation.

from cinefractal import FocalSystemInterface, FractalType, PathGenerationCriterion

focus = FocalSystemInterface(max_iterations=300)
focus.set_fractal_type(FractalType.julia())
focus.set_fractal_seed_value(-0.8, 0.156)
focus.set_extension(2.0)                 # search radius
focus.set_evaluation_extension(1e-4)     # render extension the points are judged at

result = focus.create_collection_of_points(5)
path = result.get_path_from_start_point_with_distance(
    PathGenerationCriterion.short_path(), -1.5, 0.0
)   # list of (real, imag, cumulative_distance)

Note that the search extension (where candidates are scattered) is decoupled from the evaluation extension (the extension you will actually render at; smaller = deeper zoom). The cumulative distance returned by get_path_from_start_point_with_distance is handy for timing a constant-speed camera travel — see julia_autofocus_movie.py.

Examples

All examples live in cinefractal/python/:

• iterationfield.py — query a raw iteration field and visualize it with plotly.
• mandelbrot.py — render a Mandelbrot image and save it to a file.
• julia_zoom_movie.py — a pre-scripted Julia zoom/pan rendered to MP4 with MoviePy.
• julia_autofocus_movie.py — autofocus picks interesting points and a movie tours them.
• sea_horse_zoom.py - generates the animated Gif from above.

Building from source

The project is a Cargo workspace; the Python package is the cinefractal crate. From cinefractal/:

# 1. (Re)generate the type stub cinefractal.pyi from the Rust docstrings
cargo run --bin stub_gen

# 2. Build and install the extension module into the current environment
maturin develop --release

To build the HTML documentation (Sphinx), from cinefractal/docs/:

./make.bat html        # Windows
# or: make html        # Linux/macOS

Project structure

This repository is a Rust workspace with three crates:

• fractal_core — the core fractal, animation, and focal-point engine (pure Rust).
• fractal_viz — a small native macroquad demo used to test the core without Python.
• cinefractal — the PyO3/maturin bindings that expose the engine to Python.

License

Licensed under the MIT License. Copyright (c) 2026 Christoph Lürig.