LoginSign Up
← Back to all sparks
I

InvokeAI

AI-ASSISTANTS
Velocity6.3

Open-source generative AI image creation and workflow studio

invoke.com ↗

InvokeAI adds video generation and multi-GPU, moving past its image-only boundary

image-generationvideo-generationmulti-gpumodel-supportlocal-first
◆Current state
The 6.14.0 release candidate is the largest change in this window by a wide margin: video generation via Wan 2.2 covering text-to-video and image-to-video, multi-GPU support, and a batch of new model families — Krea.2-Turbo and Raw, Ernie Turbo, Ideogram 4, plus Anima controlnets and inpainting. Everything before it is stabilization: 6.13.5 and its release candidate were explicitly maintenance cuts, 6.13.6 fixed a single crash in the Qwen Image models, and 6.13.0 in early June broadened model coverage including remotely hosted providers such as GPT Image.
◆Where it's heading
The release notes have been signposting 6.14.0 since June — the 6.13.5 notes list video generation, multi-GPU, pressure-sensitive canvas, HiDiffusion and workflow-to-workflow calls as what was coming — so the last two months read as a deliberate quiet period holding the tree stable while the feature branch landed. The direction it lands in is broadening on two axes at once: modality, from images to video, and hardware, from one GPU to several. The 6.13.0 addition of remotely hosted models points the same way, toward a workbench that orchestrates whatever backend a model needs rather than one that only runs locally.
◆Prediction
With 6.14.0 still at release candidate 1 and carrying this much new surface, the near-term work is most likely candidate iterations and fixes against the video and multi-GPU paths before stable. Several features named in the 6.13.5 notes — pressure-sensitive canvas, HiDiffusion, workflow-to-workflow calls — do not appear in the candidate's own highlights, so they are the obvious candidates for what follows.

◆Recent moves

  1. 20h ago

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

    ⚡ SPARK

    The 6.14.0 candidate delivers what the last two months of maintenance releases were holding the tree for: video generation through Wan 2.2, multi-GPU support, and new model families including Krea.2, Ernie Turbo, Ideogram 4 and Anima controlnets. It takes InvokeAI out of the image-only category it has occupied.

    View source ↗
  2. 26d ago

    Patch fixing Qwen Image crash from 6.13.5

    A single-issue patch correcting a dictionary key error introduced in 6.13.5 that crashed Qwen Image and Qwen Image Edit with the default single-file Qwen 2.2 VL Encoder. Regression cleanup during the hold before 6.14.0.

    View source ↗
  3. 27d ago

    Maintenance release ahead of 6.14.0

    A maintenance release focused on bug fixes and stability, whose notes are mostly a preview of what 6.14.0 would carry. Explicitly a holding release rather than a feature one.

    View source ↗
  4. 1mo ago

    Release candidate for the 6.13.5 maintenance cut

    The release candidate for the 6.13.5 maintenance cut, with the same content as the stable release four days later. A staging tag rather than a distinct change.

    View source ↗
  5. 1mo ago

    InvokeAI 6.13.0 adds Qwen Image and remotely hosted providers

    Broadens model coverage in both directions — new locally-hosted models including the Qwen Image family, and support for models hosted remotely by external providers such as GPT Image. The first sign in this window of InvokeAI acting as a front end for backends it does not run itself.

    View source ↗
  6. 2mo ago

    Release candidate 3 for 6.13.0

    The third release candidate for 6.13.0, carrying the same new-model content that shipped stable three weeks later. Pre-release staging.

    View source ↗