●Change8

v5.12.0rc1

📦 invokeaiView on GitHub →
✨ 3 features🐛 2 fixes⚡ 1 deprecations🔧 1 symbols

Summary

This release introduces support for Nvidia 50xx GPUs via a PyTorch update, adds model relationship management for linking LoRAs, and provides new IP Adapter methods. Internal changes were also made to improve stability regarding pydantic validation errors.

Migration Steps

  1. The new Invoke Launcher is the recommended way to install, update, and run Invoke. Follow the Quick Start guide to use it.
  2. If performing a headless install, follow the manual install guide.

✨ New Features

  • Support for Nvidia 50xx GPUs enabled by PyTorch v2.7.0 bump.
  • New model relationship feature allowing models (primarily LoRAs) to be linked to a main model in the model manager tab.
  • New IP Adapter methods: \`Style (Strong)\` and \`Style (Precise)\`.

🐛 Bug Fixes

  • Fixed GGUF quantization issues on MPS.
  • Addressed pydantic deprecation warning related to accessing \`BaseModel.model_fields\`.

🔧 Affected Symbols

GGMLTensor

⚡ Deprecations

  • The previous IP Adapter style method is renamed to \`Style (Simple)\`. Users should update to use the new explicit naming if they relied on the old default.