GPUResize

Fast and advanced gpu-based image interpolation.

GPUResize is a GPU-accelerated plug-in for Adobe After Effects and Premiere Pro with real-time performance and advanced image interpolation abilities, which are not found natively.

We believe that there’s no such thing as one and only perfect image scaling algorithm – instead, we are introducing several different ones, that allow you to make the best choice for your footage.

Use cases of GPUResize aren’t only limited to simple SD to HD video upscaling, but instead could vary from pixel-sharp upscaling of animated pixel art to downscaling your 4k and 8k camera footages while keeping the most detail possible.

Interpolation filters

• Nearest (pixel-exact)
  Performs nearest neighbor image scaling. It is similar to ‘Draft’ layer quality setting in After Effects, excepting you don’t have to care about layer quality setting anymore, it will always keep your pixels sharp. It could be used for upscaling of pixel art and ’gif style’ animations.
• Bilinear (cheap, classic)
  Performs bilinear image scaling (same as Best layer quality in all versions of After Effects prior to CC).
• Triangle (cubic, smooth, triangle shaped)
  Cubic triangle filter, the weighting decreases with increasing distance from the pixel. It provides slightly better, yet smoother results than the usual bilinear filter.
• Bell (cubic, smooth, bell shaped)
  Smooth cubic filter with Bell-curve.
• B-Spline (smooth spline filter)
  This cubic filter uses B-spline interpolating functions instead of classic cubic splines, which in general yield quite smooth results.
• Catmull-Rom (cubic, very sharp)
  Sharpening cubic filter with Catmull-Rom spline interpolation.
• Mitchell (cubic, neutral)
  Mitchell-Netravali’s separable cubic filter, good for downsampling real-world images. Very precise filter without blurring yet without much ringing/sharpening.
• Lanzcos3/Lanzcos4/Lanzcos6 (HiQ, sharp)
  Lanzcos (pronounced “LAHN-tsosh”) is named after a Hungarian mathematician and probably the best filter in terms of detail preservation and sharpness while resampling ‘usual’ real-world images. Lanzcos3/Lanzcos4/Lanzcos6 are performing filtering with the size of the window equal to 3,4 and 6 pixels respectively.
• Kaiser (HiQ, smooth)
  Kaiser-Bessel window filter; very good results, but smoother than Lanzcos.

Multipass options

• Multipass downscale
  Does exactly what it says on the tin. Mipmap style filtering, use it for big downsampling ratios (more than 2 times).

• (NEW) Stairstep upscale 
  Added by popular demand; interpolating multiple times in small increments until the goal size is reached; you can use it with any interpolation filter, but prepare to wait a bit, when you’ll be upscaling from 512x to 8k using complex filters like Kaiser/Lanzcos with this option. Generally we would advise to try it also  with simpler cubic image filters – such combo can just shine when used with several complex image types.