aaronfeng
patreon
[News] Automatic model selection for RIFE frame interpolation engine
🕑 Added 2024-02-28 07:28:13 +0000 UTCAlready available in the latest version
-----------
The new FAST & High quality RIFE model in 🔗my last post will use 1.7x more VRAM compare to the previous default v4.6 model, and it could casue low-end GPU with less VRAM unable to process high-res videos
So this NEW [Automatic RIFE model selection] feature will be available in the next update along with the new rife models, and it will be enabled by default.
When enabled, waifu2x will automatically switch to NEW High quality model if your GPU have enough VRAM, or switch back to the older v4.6 and v4 model if not.
Here is an example: when the video is 1080p and your GPU have enough VRAM, waifu2x will switch to the High quality model, when the video is 4k and your GPU doesn't have enough VRAM for the new model, waifu2x will automatically fall back to 4.6 or v4 model, which requires less VRAM, so you can still process that video.
This new feature will be available in the next stable update, thank you for continuing to support this project! 💕
Comments
Aaron Feng
That's probably because you are using 4090 and it has way more bandwidth than 4060, so the bandwidth could be the bottle neck on my 4060 and that's why we got different test results
Giovanni Santi
I'm try the new interpolation mode RIFE 4.14 In my PC (I9 13900K RTX4090) the time is the same of RIFE 4.6 The only dofference is before use 16 GB of VRAN and now with 4.14 using 19 GB. Ciao Aaron and thank you :)
Giovanni Santi
GREAT !
