v5.12.0rc2
Breaking Changes📦 invokeaiView on GitHub →
⚠ 2 breaking✨ 3 features🐛 3 fixes⚡ 1 deprecations🔧 5 symbols
Summary
This release introduces support for Nvidia 50xx GPUs and a new model relationship feature for linking LoRAs. It also enforces stricter validation on node default values, which may require updates for custom node developers.
⚠️ Breaking Changes
- Default values for node fields are now strictly validated upon application startup. If a node defines a field with a default value that does not match the field's type annotation (e.g., providing `None` for a non-optional `ImageField`), the application will error on startup. To fix, either update the type annotation to include `| None` (e.g., `ImageField | None`) if the field is optional, or remove the invalid default value or provide a valid one.
- The previous IP Adapter style method has been renamed to `Style (Simple)`. Code relying on the old name for the simple style method must be updated to use `Style (Simple)`.
Migration Steps
- If you are a node developer, review all custom nodes to ensure default values strictly adhere to their type annotations. Update optional fields to use `| None` or remove invalid defaults.
- If you use the IP Adapter style method in scripts or configurations, rename references from the old style method name to `Style (Simple)`.
✨ New Features
- Support for Nvidia 50xx GPUs added via PyTorch v2.7.0 bump.
- New model relationship feature allowing users to link models (primarily LoRAs to main models) in the model manager.
- New IP Adapter methods: `Style (Strong)` and `Style (Precise)` added.
🐛 Bug Fixes
- Fixed GGUF quantization issues on MPS devices.
- Addressed internal invocation model changes aimed at reducing `ValidationError` occurrences.
- Improved parsing and error handling for node defaults.
🔧 Affected Symbols
ImageFieldBaseInvocationInputFieldGGMLTensorBaseModel.model_fields⚡ Deprecations
- Accessing `BaseModel.model_fields` is deprecated; internal code has been updated to address the warning.