This example implements loading an ONNX based AI model using the ONNX Runtime. It uses CUDA to execute the model.
- Requires TouchDesigner 2025.32302 or later, installed via the Web Installer. Select 'Public ONNX' as the installer option. This is due to a conflict with the onnx-runtime.dll version Kinet Azure requires.
- Requires an Nvidia GPU. Currently this example uses CUDA as the execution engine. This example could be changed to use a different engine.
- Requires the CUDA 12.8 Toolkit to be installed.
- It requires ONNX runtime version 1.23.2, for CUDA 12.
- The project currently looks for the headers/libs for ONNX in C:/Program Files/onnxruntime
- Download 1.23.2 runtime from the ONNX Github. Unzip this into C:/Program Files/onnxruntime so the directories exist with the .lib and .h files:
- C:/Program Files/onnxruntime/lib
- C:/Program Files/onnxruntime/include
- The .dll files from this are not used. The .dlls that come with TouchDesigner will be what is loaded.
- It requires the candy.onnx model, which is available on the ONNX Examples Github. The model is directly available here.
- Place candy.onxx next to the .dll generated by this project that you load into the CPlsuPlus TOP (or your Plugins directory where you place the .dll).
This TOP takes one input which must be 8 bit RGBA, 720x270 resolution.