Change8

v6.14.0-rc1

📦 invokeaiView on GitHub →
16 features🐛 3 fixes🔧 19 symbols

Summary

This release introduces significant new features including video generation, support for several new models like Krea.2-Turbo and Ideogram 4, and multi-GPU capabilities. It also brings performance improvements and enhancements to the Canvas and workflow management tools.

Migration Steps

  1. Configure InvokeAI for multi-GPU generation if you have two or more GPUs installed.
  2. Select the VAE in the Model Manager to run VAEs on the CPU.

✨ New Features

  • Video generation support via Wan 2.2 (text-to-video, image-to-video, image-to-image-video).
  • Support for Krea.2-Turbo, Krea.2-Raw, Ernie Turbo, and Ideogram 4 models.
  • Anima controlnets and inpainting support.
  • Multi-GPU support for parallel generation across multiple GPUs.
  • Round robin job scheduling in multiuser mode.
  • Ability to call a saved workflow from another workflow.
  • LoRA collection picker node for stacking multiple LoRAs.
  • Improved VRAM consumption estimates for multiple models.
  • VAEs can now be run on the CPU, reducing VRAM contention.
  • Support for UNet-only LoRA sliders with configurable min, max, and default values.
  • Pressure-sensitive brush opacity on tablets and touch screens in Canvas.
  • Fine-grid hint added to move tool in Canvas.
  • New transparency lock for Gradients and Shapes in Canvas.
  • Clip Strokes extended to Bbox and Rect/Oval shapes in Canvas.
  • Optimized graph execution for faster complex workflows.
  • Crash-recoverable image storage maintenance for moving existing images into the active `image_subfolder_strategy`.

🐛 Bug Fixes

  • Fixed Qwen Image VAE working memory estimation to free cache room before decode/encode.
  • Remapped legacy single-file VL encoder keys under transformers 5.x for Qwen.
  • Improved image collection chaining.

Affected Symbols