Skip to content

v5.5.0

Latest
Latest
Compare
Choose a tag to compare
@psychedelicious psychedelicious released this 20 Dec 05:47
· 19 commits to main since this release
v5.5.0
948ecf9

This release brings support for FLUX Control LoRAs to Invoke, plus a few other fixes and enhancements.

It's also the first stable release alongside the new Invoke Launcher!

Invoke Launcher ✨

image

The Invoke Launcher is a desktop application that can install, update and run Invoke on Windows, macOS and Linux.

It can manage your existing Invoke installation - even if you previously installed with our legacy scripts.

Download the launcher to get started

Refer to the new Quick Start guide for more details. There's a workaround for macOS, which may not let you run the launcher.

FLUX Control LoRAs

Despite having "LoRA" in the name, these models are used in Invoke via Control Layers - like ControlNets. The only difference is that they do not support begin and end step percentages.

So far, BFL has released Canny and Depth models. You can install them from the Model Manager.

Other Changes

Enhancements

  • Support for FLUX Control LoRAs.

  • Improved error handling and recovery for Canvas, preventing Canvas from getting stuck if there is a network issue during some operations.

  • Reduced logging verbosity when default logging settings are used.

    Previously, all Uvicorn logging occurred at the same level as the app's logging. This logging was very verbose and frequent, and made the app's terminal output difficult to parse, with lots of extra noise.

    The Uvicorn log level is now set independently from the other log namespaces. To control it, set the log_level_network property in invokeai.yaml. The default is warning. To restore the previous log levels, set it to info (e.g. log_level_network: info).

Fixes

  • Image context menu actions to create a Regional and Global Reference Image layers were reversed.
  • Missing translation strings.
  • Canvas filters could execute twice. Besides being inefficient, on slow network connections, this could cause an error toast to appear even when the filter was successful. They now only execute once.
  • Model install error when the path contains quotes. Thanks @Quadiumm!

Internal

  • Upgrade docker image to Ubuntu 24.04 and use uv for package management.
  • Fix dynamic invocation values causing non-deterministic OpenAPI schema. This allows us to add a CI check to ensure the OpenAPI schema and TypeScript types are always in sync. Thanks @rikublock!

Translations

Installing and Updating

As mentioned above, the new Invoke Launcher is the recommended way to install, update and run Invoke. It takes care of a lot of details for you - like installing the right version of python - and runs Invoke as a desktop application.

Follow the Quick Start guide to get started with the launcher.

Legacy Scripts (not recommended!)

We recommend using the launcher, as described in the previous section!

To install or update with the outdated legacy scripts 😱, download the latest legacy scripts and follow the legacy scripts instructions.

What's Changed

New Contributors

Full Changelog: v5.4.3...v5.5.0

Contributors

weblate, ebr, and 8 other contributors
Assets 5
Loading