●Change8

v6.0.0rc3

📦 invokeaiView on GitHub →
✨ 15 features🐛 21 fixes🔧 4 symbols

Summary

Invoke v6.0.0rc3 introduces major UI/UX improvements including new Launchpads, a dedicated Generate tab, and enhanced Canvas features like PSD export and composition guides. This release also adds support for FLUX Kontext and OMI LoRAs, alongside numerous bug fixes.

Migration Steps

  1. The Invoke Launcher is the recommended way to install, update, and run Invoke.
  2. If using the launcher, follow the Quick Start guide: https://invoke-ai.github.io/InvokeAI/installation/quick_start/
  3. If performing a headless install, follow the manual install guide: https://invoke-ai.github.io/InvokeAI/installation/manual/

✨ New Features

  • Launchpads introduced for guided, intuitive interfaces across different Invoke sections.
  • New 'Generate' tab for simple Text-to-Image generation.
  • Updated layout featuring a tabbed Image Viewer.
  • Gallery now displays all images in a single grid without pagination.
  • Composition guides (e.g., Rule of Thirds) added to the bounding box on the Canvas.
  • New 'Edit' button in the image viewer to create a new canvas with the selected image.
  • Hotkeys added: Toggle all non-Raster layers on/off with 'Shift + H'.
  • Reference image management updated and moved to the left-hand settings panel.
  • Export Canvas functionality to PSD (Photoshop) format.
  • FLUX Kontext Support added to Generate, Canvas, and Workflows tabs.
  • OMI LoRA Support and additional AI Toolkit LoRA support for FLUX.
  • Prompt Expansion feature introduced (Pro Only) for AI-powered prompt enhancement and image-to-prompt conversion.
  • Workflow and upscale launchpads added.
  • Automatically select brush tool on edit.
  • Toggle for reference image visibility added.

🐛 Bug Fixes

  • Fixed names of unpublishable nodes.
  • Ensured each invocation overrides _original_model_fields with own field data.
  • Fixed 1:1 ratio for flux kontext.
  • Fixed regional guidance reference image not selecting correctly.
  • Fixed LoRA picker to default to the current base model architecture.
  • Fixed issues related to waiting for Image UX.
  • Prevented NaN from entering konva internals.
  • Fixed issues with graph building/enqueuing logic, specifically preventing invalid edges during node collection prep.
  • Fixed related models display in the new model picker.
  • Fixed Upscale creativity bug.
  • Fixed navigation on 'Edit/View workflow' buttons to point to the correct panel.
  • Fixed disabled tabs from loading.
  • Fixed regional guidance tooltip reference.
  • Handled flux-kontext reference images correctly when switching base models.
  • Fixed exception handling in the session queue API.
  • Fixed button functionality in the workflow tab.
  • Removed Canvas Session button.
  • Fixed translation reference in models picker.
  • Fixed save to gallery image update issue from the staging area.
  • Fixed Canvas session reset when a queue item is canceled.
  • Fixed brush layer positioning in PSD export.

🔧 Affected Symbols

flux_text_encoder.pyflux_denoise.pyT5TokenizerT5TokenizerFast