Choose a Model
Vapourkit ships several model families. Choose one based on the kind of video you are processing; the exact filenames and metadata are maintained in the desktop repository’s bundled model configuration.
Video models (VSR)
Section titled “Video models (VSR)”Temporally aware models for video. They use neighbouring frames to improve stability over time.
- AniRestore TFDAT - LQ anime or cartoons, including dot crawl and rainbowing
- AniRemaster TSPAN - classic anime
- AnimeUpV2 TSPAN - low-quality anime
Image-based models
Section titled “Image-based models”Frame-by-frame models. They work on video but may show shimmer or flicker on fine detail.
- AnimeJaNai HD V3 - modern anime
- AnimeJaNai HD V3 Sharp 1 - modern, higher-detail anime
- AnimeJaNai SD V1 - classic high-quality anime
- AnimeJaNai V2 - modern anime
- AniSD AC SPAN - classic high-quality SD anime
- AniSD DC SPAN - classic low-quality SD anime
- AnimeSharpV4 Fast - low-quality anime
- Animefilm bundle - low-quality SD anime
The model picker shows the full model name, scale, precision, and display tag where available.
Backend compatibility
Section titled “Backend compatibility”The included ONNX models can be used with the backend available for your platform:
| Backend | Platform | Model behavior |
|---|---|---|
| TensorRT | Windows and Linux NVIDIA systems | Builds a GPU-specific engine on first use; later runs reuse it. |
| DirectML | Windows | Runs the ONNX model directly on a DirectX 12 GPU. |
| NCNN Vulkan | Windows and Linux | Runs the ONNX model directly through Vulkan. |
The model, output resolution, precision, and inference stream count all affect memory use. The 6 GB VRAM recommendation is a starting point rather than a guarantee; reduce streams or process at a lower resolution when a model runs out of memory. TensorRT engines are tied to the GPU and software environment that built them.
Model support
Section titled “Model support”Vapourkit supports models accepted by vs-mlrt, subject to the selected backend’s requirements. See the vs-mlrt wiki for model-specific details.
For licensing information, see Model Licensing. Check the license of any model before redistributing it.
Where models live
Section titled “Where models live”At runtime, bundled and imported models are stored in data/models/ under Vapourkit’s application data directory. On Linux, the default data directory is under ~/.config/vapourkit-gui/data/.
For bringing your own model, see Use Your Own Model.