VRCFaceTracking news & patch notes

Hey all!

Another small patch today:

Improvements

• Fix head rotation parameters not being correctly passed back to the main VRCFT process by module processes

• Further module sandboxing teardown reliability by utilizing a more aggressive cleanup strategy

• Fix settings save locking issues with a debounce. This should help reduce crashes caused by race conditions inside settings saves

Features

• Add eyelid blend slider to unified correctors. This allows you to set a value by which both eyelids are blended with each other. This should assist users who experience uneven/asymmetric eyelid tracking data being provided by their hardware

Have a great rest of your week!

April Update

Sorry for the slight delay between updates!

Here's what's coming in this (slightly) early April update.

Fixes

• Head tracking data was being ignored

• Head tracking data parameters were using incorrect ranges

• Made module process termination more reliable

• Restored functionality of EyeBrow/EyeWiden expressions for older headsets

Features

• Hehe :)

Happy January

We've been working hard on new updates to improve and fix previous versions of VRCFaceTracking!

Updates

• Settings now save immediately, as well as when the app shuts down. This way, when you modify a setting and then do not close using the window controls, VRCFT still remembers.

• Added a flag inside module descriptors that ensures locally installed modules do not get overwritten by the module registry (in cases where you're using an unpublished edit of an official module.)

Fixes

• Fixed an issue during calibration where parameters would be over-driven and stuck.

• Further changes to the module process shutdown flow. Hopefully this reduces the likelihood of processes lingering upon exit.

• Fixed a crash while handling malformed VRChat avatar config JSON.

• Added proper SIGTERM listener to module processes to allow for a cleaner shutdown.

If you had migrated to a previous version of the app, we urge you to try this latest build again to see if it resolves your issues.

If you still run into issues, please provide a detailed report in the VRCFT Discord!

Thanks again for your continued bug reports and your support.

Seasons Greetings!

Hi everyone! We hope you've been enjoying the holidays! Here's some small fixes pushed since the last update in December that we'd like to release before the new year rolls around.

Changes

Calibration

We heard you loud and clear regarding the last update and so we've made some changes to the way calibration works. Namely, reverting the experimental non-linear calibration methods and disabling calibration by default in the meantime until we have a better system in the future. We apologise for any disturbance this may have caused you.

Fixes

• Fixed issue in copy and save-to-file functionality in the output tab causing log lines to be serialized to string incorrectly.

• Fixed a cross platform issue where the avatar config file would not be discovered correctly on non-windows systems

• Added more failsafes to the module sandboxing process that should cut down on the likelihood of module processes lingering after closing VRCFT

• Fixed module hardware preview images not being displayed correctly on home page

Thanks for sticking around, and we'll see you in the New Year!

Woah, a new stable update for VRCFaceTracking?? Insane. There's two main features in this update you'll care about, one of which you'll notice, and the other you won't: Calibration and Sandboxing

Calibration

Calibration is our attempt to improve uniformity across different hardware and faces. It normalizes the range of the expressions based on the data received. Put simply, a smile from one person on one headset should look the same as a smile from another person on a different headset on the same avatar.

What does this mean for me?

Avatar makers: If you 'tune' your face tracking based on your testing, now is the time to not do that. '1' is now truly what the maximum should be. Our goal is to have a fully uniform set of expressions. Before calibration, an avatar that looked correct on one headset might not look right on another.

Normal people: It should be easier to pull funny faces without diminishing other expressions. If you have a hard time frowning (me) then you should be able to do that easier. It's on by default, and there are also other settings you can adjust regarding tracking and the calibration pipeline. Data filtering is pretty neat too! Give it a try if you want a smoother look!

Sandboxing

You shouldn't notice this, but it is pretty neat.

Sandboxing improves a lot of the stability related issues regarding modules. Instead of running modules inside the same process as VRCFT, they now run in a separate little box. The main improvements are:

• Modules can no longer crash VRCFT on their own

• When using multiple modules, they can no longer clash with each other

• The pause button works now (probably). ːsteamhappyː

Brand Guide

Final thing: If you want to show off your VRCFT support, we made neat badges to do so! You can see them here! Banner Avatar Credit: Masc Canine - Julia Winterpaw

TLDR; If you're crashing, please switch to beta.

Fixes

• A recent Windows 11 update (KB5066835) created a bug in HTTP2 that causes VRCFaceTracking to crash after connecting to the VRChat OSCQuery HTTP handshake server. Microsoft currently is aware of a bug with HTTP2 and local connections.

• There is a temporary fix in the beta branch. You may need to restart VRCFaceTracking when changing avatars.

Funny Side Note

• If you've never explored it, the beta branch includes more features that haven't been pushed to the stable branch. There's new settings that allow you to tweak how your headset tracks your pretty little face.