●Change8

v6.13.0.rc3

📦 invokeaiView on GitHub →
✨ 13 features🐛 1 fixes🔧 6 symbols

Summary

InvokeAI v6.13.0 RC3 introduces extensive support for new locally-hosted models (Qwen, Anima) and integration with external API providers like OpenAI and Google. This release also overhauls the workflow editor, adds LLM-powered prompt tools, and brings significant improvements to the Canvas and Model Manager.

Migration Steps

  1. To use prompt expansion or image-to-prompt features, install a causal LM model (e.g., Qwen/Qwen2.5-1.5B-Instruct) via the HuggingFace tab in the Model Manager.
  2. To run custom workflows on a Canvas image, create a workflow with an image input field in the Form Builder and a Canvas Output node, then right-click the canvas and select 'Run Workflow'.
  3. To use workflow connectors, right-click on the canvas and select 'Add Connector'.

✨ New Features

  • Added support for locally-hosted Qwen Image (txt2img and edit versions) and Anima models.
  • Introduced support for externally-hosted models from providers like OpenAI, Gemini (Google), BytePlus, and Alibaba Cloud, requiring API keys.
  • Implemented LLM-powered prompt expansion (using a local TextLLM model) and Image-to-Prompt features accessible via new buttons in the prompt textfield.
  • Image Gallery now supports automatic creation of "virtual" boards to group images, initially by date.
  • Canvas improvements include the ability to hide/show preview stage thumbnails and the addition of a Lasso Tool (freehand and polygon modes) for masking/selection.
  • Added functionality to apply custom workflows directly to a Canvas image via a right-click context menu option.
  • Introduced a Custom Node Manager tab for installing and managing node packs (groups of related nodes).
  • Workflow nodes are now organized by function in an easy-to-navigate node picker.
  • Added the workflow connector, a new edge type for flexible connection between multiple input/output fields.
  • New generation nodes for running external models, new denoising nodes for Qwen Image and Anima local models, and a node for decoding invisible watermarks.
  • The Collect node now supports chaining for deterministic iteration.
  • A new 'If' mode allows for conditional branching in workflows.
  • In Multiuser Mode, users can create private, shared, and public image boards, and private/shared workflows.

🐛 Bug Fixes

  • Corrected many small bugs and performance bottlenecks in image generation.

Affected Symbols

CanvasWorkflow EditorModel ManagerImage GalleryCollect nodeIf mode