v5.13.0
📦 invokeaiView on GitHub →
✨ 10 features🐛 5 fixes🔧 1 symbols
Summary
This release introduces advanced Inpainting mask controls with noise and denoising limits, alongside performance improvements and better handling for SDXL aspect ratios. Users are encouraged to use the Invoke Launcher for installation and updates.
Migration Steps
- Users are recommended to use the Invoke Launcher for installation, updating, and running Invoke.
- If performing a manual install, follow the [manual install guide](https://invoke-ai.github.io/InvokeAI/installation/manual/).
✨ New Features
- Added advanced per-mask settings for Canvas Inpaint Masks, including 'Noise Level' (adds image-space noise using the global seed) and 'Denoise Limit' (caps denoising amount per mask).
- Aspect ratio selection now gives special handling to SDXL's trained sizes to reduce artifacts.
- Improved Canvas scroll-to-zoom handling, featuring smoother scaling on touchpads and snapping to common zoom levels.
- Added a button to pull the bounding box content into an empty Control Layer.
- Added ability to delete all images from the Uncategorized board via a right-click menu option.
- Prompt boxes now remember their size across sessions.
- Model Manager now supports installing Hugging Face (HF) repo subfolders via the HuggingFace tab.
- Faster Heuristic Resize algorithm implemented, used in 'New Layer from Image (Resize)' functionality.
- LoRA patcher is now allowed to skip unknown layers instead of erroring out.
- Updated Compel library for better handling of long prompts.
🐛 Bug Fixes
- Fixed an issue where the LoRA patcher would error when encountering unknown layers (now skips them).
- Fixed an issue related to scaling math to ensure landing on 100% consistently.
- Fixed an issue where auto-scaling moved away from SDXL training sizes.
- Fixed an issue with 'Preserve Masked Region' for new mask compositing.
- Fixed handling of invoke syntax for HF repo IDs when fetching HF model metadata.
🔧 Affected Symbols
New Layer from Image (Resize)