Releases: n00mkrad/cupscale
Releases · n00mkrad/cupscale
1.39.0
Changes:
- If a video gets dropped into the preview tab, you can either load the first frame or automatically switch to the video tab
- Loading the first frame of a video into the preview now works without having ffmpeg installed system-wide
- Added Lanczos scaling to resizing options
- Added automatic server selection (EU/US) for python download
- Added Patron list in preview window while no image was loaded
- Added option to stop Cupscale from starting in a maximized window
- Added prompt that will ask to install embedded Python if no Python runtime was found
- Better error messages for model interpolation (offline interpolation, not on-the-fly)
- Better error messages/handling for specific CUDA and NCNN errors
- Dependency Checker UI improvements (if one Python runtime is installed the other one won't be marked in red)
- Internal improvements (backporting of JSON-based Config code from Flowframes)
Installation:
- Extract the zip somewhere (preferably on an SSD), not in a protected folder like Program Files!
- Start Cupscale.exe
1.38.0 [Fix 1] - Real-ESRGAN, more options, fewer bugs
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
1.37.0 - Even more fixes
Changes:
- UI/UX Improvements for Model Selection Tree
- Fixed unhandled errors when trying to select a model if none exist in the model folder
- Fixed problems caused by switching to portable mode when previously using Appdata
- Better error handling for NCNN model converter
Installation:
- Extract the zip anywhere (Preferably somewhere on an SSD)
- Start Cupscale.exe
1.36.1 - Fixes, FP16 mode, and more
Changes:
- Added "RTX Fast Mode" (FP16) to make ESRGAN CUDA more lightweight and faster
- Portable mode is now enabled by default (similar to Flowframes)
- Fixed "Preserve Folder Structure" not working properly
- Fixed MozJpeg encoding not working (required runtimes are now included)
- Fixed problems with disk space check
- Other minor fixes and improvements
1.36.1 (Hotfix) Changes:
- Fixed embedded python installation not working
Installation:
- Extract the zip anywhere
- Start Cupscale.exe
1.35.1 - A few new options + NCNN Fix
Changes:
- Added an option to enable split depth caching for CUDA upscaling, which improves speed if your images are the same size
- Added the option "Auto-Expand Model Selection Tree" that allows you to toggle the auto-expansion of all model tree nodes
- Fixed NCNN upscaling not working if Cupscale is in a path that contains spaces
Update: 1.35.1 Changes:
- Fixed error when opening settings
1.34.0 - More alpha options & lots of fixes
Changes:
- Added an option to choose from all 3 alpha modes of Joey's ESRGAN fork
- Added an option to switch between Full and Binary (1-bit) or Ternary (2-bit) alpha
- Added an option to set the GPU ID that ESRGAN will use (for multi-GPU systems)
- Increased the retry count for post-processing files that are still locked by the python process (should fix batch/video upscaling problems)
- Fixed Overwrite mode not working
- Fixed fractional frame rates (like 12.5) being rounded to an integer after upscaling a video
- Removed esrgan-launcher and all related options as it has been made obsolete by Joey's ESRGAN fork
Known Issues:
- When batch upscaling a directory, having multiple files use the same filename (in different directories) will cause problems
Build 33 (Fix 3) - Auto-Tiling & Fixes
Changes:
- Updated Joey's ESRGAN to the latest version featuring automatic tiling. This can speed up upscaling by up to 5x.
- Fixed batch processing not working with multiple files (only with directories)
- Fixed a problem where videos without audio aren't exported correctly
- Error Handling improvements
Known Issues:
- When batch upscaling a directory, having multiple files use the same filename (in different directories) will cause problems
Build 32 - Video Improvements
Changes:
- Added option to preserve audio when upscaling a video
- Added experimental support for WEBM and AVI video formats
- Fixed a bug where the Dependency Checker would say that the python runtime is not installed, even though it is
Build 31 - Cancel Button & UI Improvements
Changes:
- Added Cancel button that will kill ESRGAN if it's running
- Web Installer now shows download percentage for each file
- Cupscale now checks if another instance is already running in the same data folder
- UI: The sidebar contents now automatically resize based on whether the scrollbar is visible or not
- Error handling for FFmpeg and Gifski encoder
Build 30 - Web Installer & Video Fixes
Changes:
- The resource installer is now web-based, making the executable 31 MB instead of 93 MB
- Fixed the H265 video option by switching to a different ffmpeg build
- Added GIF (non-animated) as compatible export format for image upscaling
- Better error handling for video upscaling (more to come - ffmpeg errors are unhandled still)
- Minor UI/UX improvements for NCNN upscaling
Testing:
- Successfully tested NCNN with video upscaling
- Successfully tested NCNN with Lite models