About InvokeAI 6.14.1 — AI Image Generation Canvas You Run Locally
Invoke, also known as InvokeAI, is a creative tool for generating and editing images with AI on your own hardware. It is built around a unified canvas where you generate, paint, inpaint and outpaint on layers, with node-based workflows and an organised gallery alongside. It runs as a local web app and is released under the Apache 2.0 licence.
This page links to the official Invoke Community Edition launcher, version 1.8.4, and to the source archive of InvokeAI 6.14.1, both on GitHub. Bineret did not build Invoke and is not affiliated with it.
How installation works
Invoke is not a single installer. The launcher is a desktop app that installs Invoke for you: it bundles uv, installs Python, creates a virtual environment, and installs the version of Invoke you choose, the latest stable release or a pre-release. It also updates and repairs the install, and updates itself. The launcher version and the Invoke version are separate numbers, which is why this page shows 1.8.4 and 6.14.1.
Launcher builds exist for Windows, Apple Silicon Macs and Linux. There is no Intel Mac launcher. Docker and a manual Python install remain as alternatives, and the older install scripts are being phased out.
Check your hardware first
Invoke runs on Windows 10 or newer, macOS 14 or newer, and Linux, where Ubuntu 20.04 and newer is well tested. From its hardware guide:
- Nvidia: SD 1.5 needs a 10 series card with 4 GB of VRAM and 8 GB of RAM. SDXL needs a 20 series card with 8 GB of VRAM and 16 GB of RAM. FLUX.1 needs 10 GB of VRAM and 32 GB of RAM. The largest models, such as FLUX.2 Klein 9B and Krea-2, want a 40 series card with 24 GB, though FP8 versions run on less.
- Apple Silicon: every M-series Mac works; 16 GB of memory or more is recommended.
- AMD: supported on Linux only. Intel Arc: Windows and Linux x86_64.
- Linux ARM64, such as a Raspberry Pi 5, runs in CPU-only mode: slow for local generation, fine for API-backed models.
Cards with less VRAM often still work, slower, and there is a low-VRAM mode.
What it does
- Unified Canvas: generation, inpainting, outpainting and brush tools on layers, plus shapes, lasso, text and gradient tools, and workflows you can run directly on the canvas.
- Workflows and nodes: a node editor for custom generation pipelines, a custom node manager and community nodes, with a guide for moving over from ComfyUI.
- Boards and gallery: organised storage with the prompt and settings kept in each image’s metadata for recall.
- Models: Stable Diffusion 1.5, 2.0, SDXL and 3.5, Flux.1 and Flux.2, Z-Image, Krea 2, Qwen Image and Qwen Image Edit, Ideogram 4, ERNIE-Image and more locally, and Nano Banana, GPT Image and Wan through APIs.
- Tools: model manager, embeddings, upscaling, and SAM and SAM2 segmentation.
- Multi-user mode for shared installs.
What 6.14.1 changed
Released on 6 September 2026. Workflows can be saved as PNG screenshots, middle-clicking gallery media can open it in a new tab, and Krea-2 LoRAs built with ai-toolkit’s DoRA magnitudes are supported. It closes a security hole in AlibabaCloud image downloads, improves VRAM reservation for VAEs, and improves memory handling at shutdown.
When to use something else
If you want maximum flexibility across video, audio and 3D as well as images, and are comfortable with node graphs, ComfyUI covers more ground. Without a capable GPU, local generation will be slow in any tool, though Invoke’s API-backed models still work.
Key capabilities
A canvas for artists
Generate, paint, inpaint and outpaint on layers, with brush, shape, lasso, text and gradient tools. Workflows can run straight on the canvas, so AI becomes part of editing rather than a separate step.
Nodes when you need them
A node editor builds custom generation pipelines, with a custom node manager and community nodes. There is a guide for people moving over from ComfyUI.
Installed by a launcher
The launcher bundles uv, installs Python, and installs the stable or pre-release Invoke you choose. It updates and repairs the install and updates itself, with no Python setup on your side.
Know your VRAM
SD 1.5 runs on 4 GB of VRAM, SDXL wants 8 GB with 16 GB of RAM, FLUX.1 10 GB with 32 GB of RAM, and the largest models 24 GB. Apple Silicon Macs work with 16 GB or more recommended.
What 6.14.1 fixed
Released 6 September 2026: workflows save as PNG screenshots, Krea-2 DoRA LoRAs load, a security hole in AlibabaCloud image downloads is closed, and VAE memory reservation improved.
Questions & Answers
Why are there two version numbers?
The launcher (1.8.4) and Invoke itself (6.14.1) are released separately. The launcher installs and updates whichever Invoke version you pick.
Is Invoke free?
Yes, InvokeAI is Apache 2.0. Models used through APIs, such as Nano Banana, GPT Image and Wan, are billed by their providers.
Will it run on my computer?
It needs Windows 10+, macOS 14+ or Linux. SDXL wants about 8 GB of VRAM and 16 GB of RAM on Nvidia; SD 1.5 runs on 4 GB of VRAM. Every Apple Silicon Mac works. AMD GPUs are Linux-only, and Linux ARM64 is CPU-only.
Is there a launcher for Intel Macs?
No, the macOS launcher is for Apple Silicon only.
Do I need to install Python?
No. The launcher bundles uv and uses it to install Python and create Invoke's environment for you.
I use ComfyUI. Can I move over?
Invoke's documentation includes a ComfyUI migration guide, and its workflow editor is node-based, so the concepts carry over.
Does the licence of Invoke cover the images I make?
Invoke's Apache 2.0 licence covers the software. What you may do with outputs depends on the licence of the model you used, so check it before commercial use.
Is Bineret connected to Invoke?
No. This page describes Invoke and links to files the project publishes on GitHub.
Tutorials
1. Check your hardware
Windows 10+, macOS 14+ or Linux. For SDXL, plan on an Nvidia 20 series card with 8 GB of VRAM and 16 GB of RAM; SD 1.5 runs on 4 GB of VRAM. Apple Silicon Macs work, with 16 GB or more recommended. AMD GPUs are supported on Linux only. The full table is in Invoke's hardware requirements.
2. Download and open the launcher
Windows: Invoke-Community-Edition-Setup-1.8.4.exe. Apple Silicon Mac: Invoke-Community-Edition-1.8.4-arm64.dmg. Linux:
chmod +x Invoke-Community-Edition-1.8.4.AppImage
./Invoke-Community-Edition-1.8.4.AppImage3. Install Invoke
In the launcher, choose where to install and which version: the latest stable release or a pre-release. The launcher installs Python and everything else it needs into its own environment, so you do not have to set up Python yourself.
If the install fails
Retry with Repair Mode, a checkbox on the Review step. If that does not fix it, open the dev console from the button at the bottom left of the launcher, run uv cache clean, and retry with Repair Mode on. Installing to a different location is the next thing to try.
4. Start creating
Launch Invoke from the launcher; it opens in your browser. Add models through the model manager, then generate on the canvas. On a card with little VRAM, read the low-VRAM guide in the documentation before loading large models.
Updating
Use the launcher to install newer Invoke versions. The launcher checks for its own updates at startup and asks before updating itself.
Support
Where support comes from
Invoke's documentation covers installation, the canvas, workflows and models, its FAQ covers common install problems, and the project's Discord is where it points people for more help. Bugs go to GitHub issues.
What Bineret covers
This page. If a download link is broken or something written here is wrong, tell us and we will fix it. We do not provide models, hardware advice beyond this page, or help with generation.
No warranty
Apache 2.0 provides the software as-is. Model downloads are large, so keep an eye on disk space in the install location you choose.
