v0.9.2 v6.14.0
umputun/ralphexAug 25, 2026 → Aug 25, 20260 days between releases
What's New
Major release introduces video generation, new image models, and multi-GPU support
Breaking Changes
- LoRA/ControlNet/IPAdapter support is rolled out in stages; not all LoRAs will currently load
- Major version jump may require configuration updates for existing workflows
New Features
- Wan 2.2 video generation with templates (text-to-video, image-to-video)
- Krea.2-Turbo and Krea.2-Raw model support
- Flux.2 Dev model support
- Ernie Turbo model support
- Ideogram 4 model support
Improvements
- FP8 support reducing VRAM usage by ~50%
- Native Intel XPU support
- Flux.2 PiD support for 4K super-resolution
- Anima controlnets and inpainting
Upgrade Notes
Upgrading from v0.9.2 to v6.14.0 is a major version jump requiring careful review of configuration and model compatibility. Note that LoRA/ControlNet/IPAdapter support is currently rolled out in stages, so verify existing workflows do not rely on unsupported models. Multi-GPU and FP8 features may require updated hardware or configuration settings.
The Two Releases
v0.9.2
Aug 25, 2026· 2 months ago
Introduced DeepSeek Harness support and various fixes for the knowledge base, coding agents, and retention processes.
Full release notesInvokeAI 6.14.0
Aug 25, 2026· 2 months ago
A major release featuring extensive new model support including Wan 2.2 video generation, Krea.2-Turbo/Raw, Flux.2 Dev, Ernie Turbo, and Ideogram 4. It introduces multi-GPU support, FP8 optimization for reduced VRAM usage, and various workflow and canvas improvements.
Full release notes