SwiftVideoSR is a Mac App which is available on Mac App Store. You can find the App by searching for 'swiftvideosr' in Mac App Store, and the App only supports Mac M series chips.
SwiftVideoSR can perform super-resolution upscaling on both images and videos, supporting both anime and realistic styles, as well as batch export. Images and videos will not be uploaded to a server; the App runs locally, thus providing very fast speed.
SwiftVideoSR includes over eighty models trained from scratch, so users can always find a suitable model for their images and videos to achieve best repair results.
The user interface is very intuitive, and one can learn to use it almost without study, which is why in this tutorial, I will only introduce some key points.
Newbie mode is suitable for those who do not want to adjust any parameters. Just import an image or a video, and export directly, as the App will automatically detect the haze level of the image or video and try to generate best result. If you feel that the automatically detected haze level is not enough or is too much, you can switch to fine-tuning mode to make slight adjustments based on the automatic detection for better results.
Expert mode includes a lot of parameters that you can explore to feel their impact on the picture. However, the most important parameter is lens blur; I usually find that adjusting the lens blur parameter is sufficient.
If you want to repair films or TV shows, I strongly recommend the 'Suitable for Movies & TV Shows' mode, because all models in this mode are trained for movies and TV shows. This allows for recovering picture details with cutting-edge generative adversarial technology when enlarging video dimensions, and the results are impressive, I personally think that the result is even superior to Topaz Video AI.
For images and videos that are already relatively clear, you can choose the 'Suitable for Clear Images' and 'Suitable for Clear Videos' modes. These two modes are designed specifically for enhancing details in pictures that are already clear, allowing for better recovering of details when upscaling clear images and videos.
Generative image enhancement utilizes diffusion models for image detail enhancement; however, the current version only performs well on clear images, while its result on blurry images is lacking and needs improvement.