v0.5.5

Update to Pytorch 2.7 in order to support Nvidia 50 series GPUs

v0.5.4

Bump the Pytorch version to 2.5.1. This seems to fix the issue many people were having with the propagation not working.

v0.5.3

This version specifies specific versions of dependencies to be installed, rather than just installing the latest versions (which previously caused the application to break). There are no further changes.

v0.5.2

Update the install_dependencies script to require pytorch version 2.3.1. The newer version 2.4 causes issues.

v0.5.1

Changelog:
This is a bugfix release to fix some major issues that prevented the application from working.

• Added 'requests' to requirements.txt. This was previously installed implicitly through another dependency, but that doesn't seem to be the case anymore.
• Fixed an issue with how the configuration file was loaded, which prevented the application from launching in CPU mode.

v0.5.0

Changelog:

• Added undo support
• Ability to import multiple masks at once
• When importing masks, it will look for numbers in the filename to interpret as the frame number that the mask should be loaded on
• Added a list of frames that are saved to the permanent memory
• Framerate in export settings will automatically populate from the input video
• Small tweaks and bug fixes.

v0.4.7

Changelog:

• When out of VRAM, it will now only clear the torch cache instead of clearing all memory. This should no longer have a negative effect on the output quality.
• Propagation no longer updates the frame it starts from, unless you start propagating on the first or last frame of the clip. This fixes issues with low quality masks appearing in the middle of the clip.
• Fixed a race condition when importing masks that prevented the mask from loading
• Added additional settings to the configuration file:
  • Specify a file for the background layer to display in the preview
  • Disable automatic clearing of the torch cache
  • Extract frames from video as either jpg or png

v0.4.6

Changelog:

• Add checkbox to the launcher to reset the workspace
• Replaced the tips area with link to the documentation on the wiki
• Move many settings into the config file at cutie\config\gui_config.yaml
• Tighten up the refined masks so the transition area is smaller
• Fixed a bug where refined masks didn't work when exporting as a matte.
• Update license from GPL3 to MIT in accordance with the original Cutie repo changing its license.

v0.4.5

Changelog:

• Aniclick anime segmentation model updated. Should work much better than the previous version
• Add preview on a background color or background image
• Clear VRAM when it gets full. While not an elegant solution, it is preferable to the application freezing or crashing.
• Fixed bug in the minimap display
• Fixed tooltips color in the dark theme
• Support dragging and dropping an input video into the launcher
• Display a warning when attempting to load a long video

Aniclick V2 model release

Significantly improved version of the Aniclick interactive segmentation model. It is more accurate and also allows selecting a single character much more easily.