●Change8

v5.6.0rc4

📦 invokeaiView on GitHub →
✨ 6 features🐛 11 fixes

Summary

This release introduces significant memory management improvements, enabling Low-VRAM mode for larger model generation, alongside major enhancements to Workflow Batches with new data types and grouping capabilities. New Blur and Noise Canvas filters are also available.

Migration Steps

  1. For users wanting to enable Low-VRAM mode, add the following line to your `invokeai.yaml` file: yaml enable_partial_loading: true
  2. Windows users should also follow the guide to disable the Nvidia sysmem fallback: https://invoke-ai.github.io/InvokeAI/features/low-vram/#disabling-nvidia-sysmem-fallback-windows-only.

✨ New Features

  • Added new Blur and Noise filters to the Canvas.
  • Expanded batch capabilities in Workflows to support float, integer, and string batch data types.
  • Introduced batch data generators for floats and integers (Arithmetic Sequence, Linear Distribution, Uniform Random Distribution, Parse String).
  • Added support for grouped (zipped) batches in Workflows, allowing execution based on paired values instead of the Cartesian product.
  • Implemented Low-VRAM mode features including partial model loading, dynamic cache sizes, and working memory adjustments to support large models on low-VRAM GPUs.
  • Added `keep_ram_copy_of_weights` config option to reduce average RAM usage.

🐛 Bug Fixes

  • Fixed an issue preventing users from typing in textarea fields in the workflow editor.
  • Reduced peak memory usage during FLUX model loading.
  • Revised the default logic for the model cache RAM limit to be more conservative.
  • Fixed image quality degradation when inpainting an image repeatedly.
  • Fixed an issue where transparent Canvas filter previews blended incorrectly with the unfiltered parent layer.
  • Fixed an issue where excessively long board names could cause performance issues.
  • Fixed an error when using DPM++ schedulers with certain models.
  • Fixed (potentially) the application scrolling off screen when run via the launcher.
  • Fixed the link to `Scale` setting's support docs.
  • Reworked error handling when installing models from a URL.
  • Fixed launcher issues, including requiring repeated installation and incorrect CPU usage on AMD GPU systems (Launcher v1.2.1).