Tune enhancr: Settings
Settings are crucial to the user experience of enhancr, as they provide various options that can significantly affect your experience.
Page 1 contains UI customization options and information about your system. On this page, you can change the accent color of the UI, disable blur, enable preview, and enable Discord Rich Presence integration.
Page 2 contains options related to RIFE, interpolation, and DPIR.
For RIFE, there are two toggles available. The first toggle enables TTA/Ensemble, which improves the quality at the cost of slower speeds. The second toggle enables UHD mode, which can help with high-resolution input resolutions.
For interpolation, there are two toggles available. The first toggle enables scene detect, which scans the input for potential scene changes and avoids interpolating between the two frames where the change occurs, reducing artifacting. The second toggle enables the ability to skip duplicate frames, which speeds up inference time by skipping frames that do not contain motion. This is useful for animated content.
For DPIR, you can adjust the values for denoising or deblocking to your desired level.
Page 3 includes Engine settings, Tiling settings, and TensorRT settings.
In the Engine settings, you can adjust the shape resolution of the engine file to address shape size problems or VRAM errors. The resolution is formatted as heightxwidth rather than widthxheight. To use a custom size, enter the size of your input or the size of your tiles (if tiling is enabled) and enable the override toggle.
The Tiling settings allow you to adjust the tile resolution and enable tiling, which is the process of dividing the video into segments or "tiles" to manage VRAM constraints.
In the TensorRT settings, you can adjust the number of streams and enable FP16. The stream count is important for performance as having too few streams can underutilize your GPU, while having too many can overutilize it. For lower-end cards, use a lower number of streams, and for higher-end cards with more VRAM, use a higher number. You can continue to increase the number as long as your CUDA cores or VRAM are not maxed out. FP16, also known as half precision, can significantly increase inference times on modern cards. So enabling it is recommended if your GPU supports it.
Page 4 includes settings related to encoding, trimming, APUs, and scene change sensitivity.
In the Encoding settings, you can enable hardware encoding, which utilizes the encoders on your GPU to encode the videos rather than your CPU. This increases speed but may decrease quality and encode efficiency.
The Trim settings allow you to toggle precision trimming. When this option is disabled, your video is cut at keyframes rather than the exact timestamp you specify. With it enabled, you can cut precisely. Be aware that by enabling this your video will get re-encoded.
The option below allows you to show engines that may not be supported by your system. This is also a workaround for a bug that people with APUs may encounter, who cannot access all of the engines, even though they have Nvidia cards.
In the Sensitivity settings, you can adjust the threshold for scene detect. A lower threshold will result in more false positives, potentially making your video look stuttery, while increasing the threshold reduces false positives, but may also result in fewer scenes being detected and potentially more areas where scenes are not detected. The toggle allows you to override the default value set by enhancr.
Page 5 includes options for custom Real-ESRGAN models and some enhancr-related options.
enhancr has the ability to use custom models for Real-ESRGAN/ESRGAN and automatically converts .pth files to .onnx. Use the dropdown to choose models, toggle the option to enable the use of the selected model, and use the "Open custom models folder" button to store your custom models.
You can change the cache location directly below this section. enhancr caches files in a separate directory before they are moved to the output directory, which simplifies processing on multiple systems. You may want to change the cache location if your main drive is running low on space, for example.
Do not use system python instead of the packaged python unless you know what you are doing.
Multi-language support will be available soon:tm:
