Alpha v0.3.1

This update is just a bunch of bug fixes. Thanks to everyone who reported an issue!

Bug Fixes:

• Fixed special unicode characters (accented characters, Japanese characters, etc) for real this time
• Fixed the dependency manager not reloading properly after installing or uninstalling something
• Fixed some NCNN auto-detection stuff to allow more than just ESRGAN-based models, such as Waifu2x (however only those two kinds of models have been tested)
• Fixed a potential bug with NCNN where the scale would not be calculated properly on certain models
• Fixed an issue with NCNN model chaining where the output would look "deep-fried"
• Fixed NCNN model re-use, which was causing the backend to crash
• Fix the "double click to add node" feature
• Fix preview sizing issue on hi-dpi devices with scaling enabled
• Hide VRAM percentage if no Nvidia device is detected as this is an Nvidia-only feature
• Fixed bug in stack images node which caused it to error on running

Alpha v0.3.0

It might not seem like it, but this is my biggest update yet with a few pretty big features and changes. The biggest by far is NCNN support. If you don't know what that means, it just means that far more devices (AMD and Intel GPUs that support Vulkan) will be able to do upscaling. This is huge and puts chaiNNer closer to feature parity with other GUIs. If you're wondering why this took so long, I basically had to figure out myself (with the help of @styler00dollar) how to get Vulkan working with the official NCNN Python api, and I ended up having to publish an entirely separate package to PyPi to get it working.

The other big change in this release is a complete redesign of how nodes are named and categorized. I decided that the existing way was too technical and that I should make everything a bit more understandable for the end-user, rather than over-complicating things. So for example, nodes aren't categorized by "OpenCV" or "NumPy" anymore, but rather "Image" and "Image (Utility)", and node names are no longer "Image::Read" and "Image::Show" but rather "Load Image" and "Preview Image". You also might notice each node also now has an icon associated with it instead of each category all being the same icon. Under the hood, many of these are exactly the same, there's just a new coat of paint on top. And don't worry, old chain files (.chn) will be automatically converted to this new format.

More is still to come.

Full Changelog:

Features

• NCNN for cross-platform, cross-gpu inference via Vulkan.
• Full node redesign including new names, icons, and descriptions.
• Merger of some nodes that were either unnecessary or redundant.
• "Snap To Grid" for node positions.
• Better invalid nodes error message which tells you which nodes are invalid.
• New "Note" node for adding comments to your chains.
• Optimized a few things which mean chain files will now be significantly smaller.
• Some inputs now support both user input and connections, though they can't actually be utilized like that at the moment (still a work in progress).
• Nodes can now be double clicked to add to the editor

Bug Fixes

• Fixed duplicating nodes not copying input data.
• Fixed some node defaults not being loaded in correctly.
• Fixed file selection inputs not remembering your last file selected.
• Loading invalid chains will now warn you that it is invalid.

Alpha v0.2.1

Changelog:

Bug Fixes

• Fix a critical issue causing the CPU and FP16 toggles to not work

Alpha v0.2.0

Changelog:

Features

• Integrated Python (separate virtual env from your system python) now gets automatically downloaded and used on startup. No more having to worry about getting the right version and making sure your dependencies aren't outdated.
• Reworked the Settings panel to have separate tabs, one of which includes an option to disable the integrated Python and use system Python instead (the old default)
• Reworked the Dependency Manager to show download/install progress, as well as installing specific versions to avoid incompatibility.
• Made node connections behave similarly to Unreal Engine blueprints when positioned in certain ways.
• Revamped how slider inputs look so that they now display useful data.
• Added a check in the interpolation nodes that will error if two models are incompatible, either due to using a different architecture or from not sharing a pretrained model ancestor.
• Reworked the logic for how nodes get visually updated while running, meaning more kinds of updates besides just finished connections can be shown more easily in the future.

Bug Fixes

• Since updates have to be vetted now, and due to the new way I install pip packages, updating Pytorch will no longer install the CPU version.
• The concatenate nodes now do a proportional resize rather than a 1:1 resize when concatenating images of different shapes.
• Fixed a bug where image names that contained unicode characters would error instead of being read.
• Number inputs are now allowed to be empty, which will be marked as invalid.
• Potential VRAM improvements (YMMV, let me know if you're still getting stuck VRAM)
• Loading .chn files that were saved while running the chain will no longer import animated.

Alpha v0.1.4

Changelog

Bug Fixes

• Fixed a bug that caused numerical inputs to become NaN, and also caused certain numerical inputs to not support decimals anymore

Alpha v0.1.3

Changelog:

Bug Fixes

• Fixed a bug that caused CPU to be force-enabled for some Nvidia users
• Fixed a few bugs that caused PyTorch to not free up VRAM during the upscale process
• Fixed a bug that would prevent the backend python process from being killed correctly
• Fixed FP16 and CPU modes not being mutually exclusive
• Fixed some issues with the concatenate nodes
• Fixed some issues with the slider inputs
• Fixed an issue where processing for too long would result in a blank error
• Fixed an issue with finding an open port for the backend process which could sometimes result in the app not opening

Features

• Added more build targets (portable .zips for Windows and Linux, DMG for MacOS)
• Added new Blur and Shift nodes as well as more color convert options (thanks @FlashlightET)
• Added a delete button for connections to make them easier to remove
• Added a default for numerical inputs (defaults to 0 now)
• Expanded the amount the user is allowed to zoom

Alpha v0.1.2

Alpha v0.1.2

Changelog:

• Fixed issue with the bundled server where certain dependencies would not be installed. Now the server is no longer bundled and instead, it silently installed the required package on startup.

Alpha v0.1.1

Alpha v0.1.1

Changelog:

• Removed an unneeded pip install that happened on startup

Alpha v0.1.0

The initial alpha release of chaiNNer