Skip to content

1.38.0 [Fix 1] - Real-ESRGAN, more options, fewer bugs
Compare
Choose a tag to compare
@n00mkrad n00mkrad released this 23 Aug 10:30
· 21 commits to master since this release
1.38
1b305e5

Changes:

  • Added support for Real-ESRGAN - Similar to ESRGAN NCNN, much faster, but currently only works with 4x scale models
  • Improved NCNN converter to work with Real-ESRGAN
  • Added a dropdown to select the AI implementation (you can now choose ESRGAN-NCNN here, replaces the CUDA Fallback option)
  • Added new options for ESRGAN-NCNN and RealESRGAN-NCNN: TTA, override Tile Size
  • Last used model is now saved and loaded after restarting Cupscale
  • Embedded python now works with Ampere (RTX 3000) GPUs
  • There is now a default model folder because some users where too overwhelmed by the model folder dialog
  • Three "default" models are now bundled: RealSR (DF2K_JPEG), BSRGAN, and Real-ESRGAN
  • GPU list in the main window now shows all GPUs, not just Nvidia cards (VRAM measurement only works with Nvidia)
  • Better error handling and more verbose error messages
  • Removed web-based installer, instead all components are bundled in the zip (apart from the Python runtime)
  • Fixed a bug where the preview would throw an error if your output is smaller than the input (e.g. 1x model with 50% resize)
  • General stability improvements & bugfixes
  • Tons of internal changes, cleanup, refactoring

Installation: (Fresh install recommended, don't overwrite 1.37 or any other version)

  • Extract the zip somewhere (preferably on an SSD), not in a protected folder like Program Files!
  • Start Cupscale.exe

Changes 1.38.0f1 (Fix 1):

  • Fixed bug where installing the embedded python runtime wouldn't automatically enable it
  • Removed obsolete settings (Resource Files)
  • Updated build date
Assets 3
Loading