A GPU renderer for prototyping and research written purely in python
BMW scene from Mike Pan
Download the zip of this repo and install through Blender's plugin mechanism. Taichi will be installed automatically the first time you load.
Alternatively, if you are a developer you probably want to manage this somewhere you edit your code and symlink the directory to your Blender's addon directory.
This is an addon for a renderer to run as an add-on to Blender. Blender has many renderers already, including the very good Cycles renderer built in, and (shameless self-promotion of other projects I'm involved in) Radeon ProRender. If you are looking to render pictures for artwork, you should use one of those.
It seems half the repos I see on github are someone's "Toy Renderer" project. Do we really need another one?
The difference with this renderer is it is targeted at the user who is interested developing rendering code. This allows you to:
The biggest question here is executing python code on the GPU. All credit there has to go to Taichi which is a very simple way to execute python code on various devices. It's not perfect (yet) but as a Python developer I found it very easy to pick up.
Currently the renderer will load and path trace scenes in Blender when you hit the render button (F12). Basic Blender material settings and lights are supported. My plan is to do plenty of optimizations (there's not even a BVH yet), and then tackle the big issue of full node-based materials.
Some known issues:
Yes please! I would very much like this to be a thing that other developers contribute to, and use in their work.
Repo structure:
__init__.py
Blender registrationui.py
Panels for drawing the renderer properties in Blender (note it currently just uses Cycles samples and ray bounces)engine.py
This is the class that Blender calls to execute the renderer, update the scene etc. It passes data to everything in the render/
directoryexport/*
The export code. Blender data is exported to numpy arrays. render/*
The rendering code. Numpy arrays from export
are passed to here and moved to taichi arrays and used for rendering on the GPU.