●Change8

v5.9.0

📦 invokeaiView on GitHub →
✨ 7 features🐛 10 fixes🔧 2 symbols

Summary

This release introduces FLUX Fill support for high-quality inpainting/outpainting and beta support for the LLaVA OneVision VLLM multimodal model. It also includes numerous UX improvements, bug fixes, and internal refactoring for model handling.

Migration Steps

  1. If using FLUX Fill, note that it is not compatible with standard Text to Image or Image to Image operations on Canvas; an inpaint mask or empty bbox regions are required.
  2. Users are recommended to use the new Invoke Launcher for installation and updates, following the Quick Start guide.
  3. Users needing a headless install should follow the manual install guide.

✨ New Features

  • Added FLUX Fill support in Workflows and Canvas for high-quality inpainting and outpainting.
  • Beta support for the LLaVA OneVision VLLM family of models, usable in Workflows with the `LLaVA OneVision VLLM` node.
  • Support for custom string field drop-downs in Workflow Builder.
  • The About modal now displays the app's runtime settings, distinguishing between defaults and user-defined settings from `invokeai.yaml`.
  • Improved UX for missing or unexpected fields in Workflows.
  • De-wonkified LoRA node names (reverting changes from v5.8.0).
  • Support for hot reload for custom nodes.

🐛 Bug Fixes

  • Fixed issue where shift-clicking to draw on Canvas ignored `Clip to Bbox` setting.
  • Fixed issue with Image Viewer where the image could overflow the viewer.
  • Fixed overflow issue with very long node titles.
  • Fixed a minor visual bug in string generator nodes.
  • Fixed SigLip model translation key issue.
  • Fixed mask logic in graph builders.
  • Restored `display: flex` to image viewer and node editor.
  • Fixed issue where dev_reload did not work for files in nodes/.
  • Made source location discovery more robust.
  • Fixed LLaVA OneVision model calculating its own size incorrectly.

🔧 Affected Symbols

LLaVA OneVision VLLM nodeinvokeai.yaml