Upscale your videos using R-ESRGAN
- Run the portable executable file or the
R-ESRGAN-AnimeVideo-UI.pyw
script to launch the UI. - Wait for the required files to downloaded. ~200mb - ffmpeg, ffprobe realesrgan-ncnn-vulkan-20220424-windows, models from this repo.
- Select a video file that you want to upscale.
- Follow the steps labeled in the UI.
Note: The Upscale
and Merge
operations delete the previous frames by default. If you want to keep the frames, make sure to enable the Keep Frames
option.
After Extracting or Upscaling, you have the option to scale the frames to any size, either by percent or exact resolution.
mp4
, gif
, avi
, mkv
, webm
, mov
, m4v
, wmv
, are supported.
- Other Uses:
- Batch Upscale a folder of images.
- Upscale a single image.
- You can also use these models for upscaling:
RealESRGAN_General_x4_3
,realesrgan-x4plus
,realesrgan-x4plus-anime
- Output any video as a gif, and/or output gif as mp4.
Why use this over an application like Chainner, or Topaz Video AI?
While both applications are undoubtedly more than capable, (among others, I'm sure) I found xinntao's implementation is extremely fast compared to Chainner. And of course free is always better than Topaz Video AI's high cost.
Other upscale models are popular, but realesr-animevideov3 remains a favorite of mine for multiple reasons.
- It's very fast. Speed Tests⏱️
- While it does have somewhat heavy smoothing, it doesn't totally destroy fine details like grain.
- In my opinion, it has superb line reconstruction compared to others like
RealESRGAN_x4plus_anime_6B
Check out these comparisons.✨
You don't need to worry about anything if you're using the portable/executable version.
You must have Python 3.10+ installed to the system PATH
Running the script will automatically fulfill all requirements.
The pillow
library will be downloaded and installed (if not already available) upon first launch.
ffmpeg
and ffprobe
will be downloaded upon first launch.
realesrgan-ncnn-vulkan-20220424-windows.zip
and the required models will be downloaded upon launch.
- New:
- A terminal window now opens with the app to display various info/errors etc.
- This is really helpful during the initial startup while downloading files, and checking previous completed processes, etc.
- An hstack video is now created during the "Create Sample" process.
- You can now set an audio bitrate.
- You can now enter an exact audio or video bitrate value.
- New preset for Audio/Video Bitrate options,
Auto From Source
.- Regular "Auto" mode allows FFmpeg to choose a bitrate when merging. This generally results in lower quality video and audio.
- When you select "Auto From Source" for "Video Bitrate", it will use source video bitrate and add 1500.
- When you select "Auto From Source" for "Audio Bitrate", it will use the source audio bitrate.
Batch Resize Image
Added to the "Image" tab. From here you can select a folder and resize images to any resolution.- There's also an option to resize a single image.
- You can now define a start time for creating the preview sample,
- Output videos are now encoded with a keyframe every 10 video frames. This results in much smoother seeking. but larger file size.
- You can now skip the upscale process. This is useful if you just want to resize a video
- A terminal window now opens with the app to display various info/errors etc.
-
Fixed:
- Video frame rate info collection is now more robust.
- Fix an issue where the app would remain running if closed during "auto > upscale"
- Handle an error when "Toggle Animated Thumbnail" doesn't work.
- Fixed error preventing multipliers like "x0.25", "x4" from working as a resize input.
- Fixed error that would occur after first selecting an invalid filetype, then selecting a valid filetype.
-
Other changes:
Batch Image Upscale
Added back to the Tools menu. This just opens the Image tab.- Renamed "R-ESRGAN-AnimeVideo-UI.pyw" to "R-ESRGAN-AnimeVideo-UI.py"
- Moved "icon.ico" to "bin" folder.
ffmpeg-6.0-essentials: https://ffmpeg.org/
Real-ESRGAN_portable: https://github.com/xinntao/Real-ESRGAN#portable-executable-files-ncnn