●Change8

v5.6.1

📦 invokeaiView on GitHub →
✨ 9 features🐛 6 fixes🔧 4 symbols

Summary

This release focuses on stability and usability improvements, including expanded LoRA compatibility, new canvas editing features, and better memory management defaults. Users are encouraged to adopt the new Invoke Launcher for installation and updates.

Migration Steps

  1. The new Invoke Launcher is the recommended way to install, update and run Invoke. Follow the Quick Start guide to get started with the launcher.
  2. If not using the launcher, follow the manual install guide for a headless install.

✨ New Features

  • Expanded FLUX LoRA compatibility, including support for FLUX OneTrainer LoRA formats (incl. DoRA).
  • On-demand model cache clearing via a button on the Queue tab.
  • Canvas Adjust Image filter (levels, hue, etc.) added.
  • Button added to cancel all queue items except the currently running one.
  • Functionality to Copy Canvas/Bbox as image via the Canvas right-click menu.
  • Functionality to Paste image into Canvas/Bbox using the normal paste hotkey, prompting for placement.
  • Allow 'Collect' nodes to be connected directly to 'Iterate' nodes.
  • Allow 'Any' type node inputs to accept collections (e.g., 'Metadata Item' node's value field now accepts collections).
  • LoRA Loader node LoRA collection input is now optional, supporting metadata nodes.

🐛 Bug Fixes

  • Fixed issues where the staging area got stuck if one image failed to load (e.g., if it was deleted).
  • Fixed bug with some LoRA variants when applied to bitsandbytes NF4 quantized models.
  • Improved MaskOutput dimension consistency.
  • Fixed T5EncoderField initialization in SD3 model loader.
  • Fixed '[object object]' error message in OOM toast notifications.
  • Improved error messages when invalid graphs are queued.

🔧 Affected Symbols

LoRA Loader nodeMetadata Item nodeCollect nodeIterate node