Use Your Own Model

Use the Import Custom Model action in Vapourkit to add an ONNX model:

1. Choose the .onnx file.
2. Enter the model name shown in the model picker.
3. Select the model type: Image or VSR.
4. Choose the inference backend and precision options shown for that backend.
5. Click Import Model.

Imported models are stored in Vapourkit’s data/models/ directory and appear in the model picker after import.

Backend behavior

Backend	Platform	Build behavior	Input file
TensorRT	Windows and Linux NVIDIA systems	Compiles a GPU-specific .engine; first use may take several minutes.	ONNX source is retained for rebuilding.
DirectML	Windows	No engine build.	ONNX file directly.
NCNN Vulkan	Windows and Linux	No engine build.	ONNX file directly.

On Linux, TensorRT also requires a compatible NVIDIA CUDA/TensorRT stack on the host.

TensorRT engine files are tied to the GPU and driver environment that built them. Keep the original ONNX file if you may need to rebuild the engine or use the model with another backend.

Supported architectures

Vapourkit supports model architectures accepted by vs-mlrt, subject to the selected backend. See the vs-mlrt model documentation.

Finding models

OpenModelDB is a community catalog of image and video restoration models.

Licensing

Check the license of any model you import. See Model Licensing for the licenses of Vapourkit’s bundled models.