LoginSign Up
← Back to all sparks
I

InvokeAI

AI-ASSISTANTS
Velocity6.3

Open-source generative AI image creation and workflow studio

invoke.com ↗

InvokeAI's video release is on its second candidate, now with Intel GPUs in scope.

image-generationvideo-generationself-hostedmulti-gpumodel-supportintel-xpu
◆Current state
6.14.0 has been in release candidates since 31 July and is the feature cut that adds video generation via Wan 2.2, multi-GPU execution, and a long list of new model families. RC2 extends the same release rather than starting a new one: Flux.2 Dev, Flux.2 PiD super resolution to 4K, and native Intel XPU support join the RC1 list. Before this train, the product spent June and early July on maintenance releases explicitly described as clearing the way for 6.14.0.
◆Where it's heading
InvokeAI is broadening on two axes at once - what it can generate, and what it can run on. The model list grows most releases, but the hardware work is the harder-won part: multi-GPU in RC1, native Intel XPU in RC2, ROCm 7.1 in the 6.13.5 maintenance cut, plus VRAM behavior fixes and idle-GPU offloading for text encoders. For a self-hosted tool, running on whatever silicon a user already owns is the constraint that decides adoption, and it is being addressed release by release.
◆Prediction
The RC series has absorbed two rounds of additions without a final tag, so expect either an RC3 or the 6.14.0 release itself next, with the pressure-sensitive canvas and workflow-to-workflow calls named back in June still outstanding.

◆Recent moves

  1. 2d ago

    6.14.0 RC2 adds Flux.2, 4K super resolution, and Intel XPU

    The second candidate for 6.14.0 restates RC1's feature list and adds Flux.2 Dev, Flux.2 PiD super resolution to 4K, and native Intel XPU support, plus HiDiffusion integration and VRAM fixes deeper in the notes. The directional move - video generation and multi-GPU - already landed in RC1, so this is the same release growing rather than a new one.

    View source ↗
  2. 18d ago

    InvokeAI 6.14.0 RC1 adds Wan 2.2 video generation and multi-GPU

    ⚡ SPARK

    The candidate that took InvokeAI from stills to video, with Wan 2.2 text-to-video, image-to-video, and interpolation between two images, delivered through the workflow editor with template workflows. Multi-GPU support in the same cut is what makes local video inference practical, and RC2 has since extended the release without changing that direction.

    View source ↗
  3. 1mo ago

    Patch fixing Qwen Image crash from 6.13.5

    A single-bug patch for a dictionary key error that crashed Qwen Image and Qwen Image Edit with the default single-file Qwen 2.2 VL encoder. Cleanup on the maintenance line before the 6.14.0 train started.

    View source ↗
  4. 1mo ago

    Maintenance release ahead of 6.14.0

    Bug fixes and library updates - React 19, ROCm 7.1, Transformers 5.5.4 - with the notes stating outright that 6.14.0 is the next feature release. The ROCm bump belongs to the same hardware-breadth thread that later produced multi-GPU and Intel XPU support.

    View source ↗
  5. 1mo ago

    Release candidate for the 6.13.5 maintenance cut

    The candidate for the same maintenance release, carrying an effectively identical body to the final. Nothing here that the 6.13.5 entry does not also say.

    View source ↗
  6. 2mo ago

    InvokeAI 6.13.0 adds Qwen Image and remotely hosted providers

    Qwen Image txt2img and edit models plus Anima land locally, and the release opens the door to remotely hosted models such as GPT Image. Adding another image model family is routine for this product; the hosted-provider path was the more interesting half.

    View source ↗