Overview
InvokeAI is a locally run application for generating and editing images with diffusion models. Unlike tools built around a single linear "type a prompt, get an image" loop, InvokeAI is organized around two production-oriented surfaces: a Unified Canvas for layered, non-destructive compositing and retouching, and a node-based Workflow Editor for assembling and reusing complex generation pipelines. The project describes itself as built for production workflows and creative professionals, which shows up in the emphasis on reproducibility, layer management, and gallery/board organization rather than raw generation speed alone.
Installation and access
InvokeAI is installed and run on your own machine via a dedicated launcher/installer, with the application then served through a local web UI in your browser. It supports Windows 10+, macOS 14+, and Linux (Ubuntu 20.04+ is the best-tested distribution). Linux ARM64 devices, such as Raspberry Pi–class single-board computers, are supported in CPU-only mode. There is no hosted version of InvokeAI itself to sign up for: the previously offered hosted service tied to the project has been discontinued, and third-party hosts offer InvokeAI deployments if a user prefers not to self-host.
Interface and workflow
The two core surfaces are the Unified Canvas and the Workflow Editor. The Unified Canvas supports independent, manipulable layers for painting, masking, and compositing generated content together, closer to a digital-painting workflow than a single-shot prompt box. The Workflow Editor is a node graph: individual generation steps (sampling, ControlNet conditioning, upscaling, and so on) are represented as connected nodes, which can be saved and reused as repeatable pipelines. A board and gallery system organizes generated images with their metadata for later retrieval.
Model and format compatibility
InvokeAI supports Stable Diffusion 1.5, SD 2.0, SDXL, SD 3.5, and Flux-family models in multiple sizes, alongside newer checkpoints such as CogView 4 and Ideogram-derived models as the project adds support. It reads common local checkpoint formats, including .ckpt, diffusers-format folders, GGUF, and safetensors. Object segmentation via SAM/SAM2 is built in for masking workflows.
Hardware requirements and performance characteristics
VRAM needs vary substantially by the model family in use: roughly 4GB for SD 1.5, 8GB for SDXL, 10GB or more for FLUX.1, and up to 24GB for larger FLUX.2/Krea-class models (some of which can run in less VRAM using FP8 quantization). NVIDIA GPUs are supported on Windows and Linux; AMD GPUs are supported on Linux only, via ROCm. Apple Silicon Macs are supported, with 16GB+ unified memory recommended. A low-VRAM mode, introduced in InvokeAI 5.6, lets systems without enough VRAM to hold a full model still generate images by offloading more aggressively, at the cost of speed.
Community and ecosystem
InvokeAI is open-source and community-developed, with an active Discord server for support and a GitHub repository accepting code, documentation, and testing contributions. Checkpoints and LoRAs used with InvokeAI commonly come from community model-sharing sites such as Civitai and Hugging Face.
Content moderation posture
InvokeAI ships with no built-in content filter or NSFW classifier. What a given session can produce depends entirely on the checkpoint the user loads and how it was trained, plus any workflow constraints the user builds themselves: the software imposes no policy of its own.
Important distinctions from adjacent tools
Compared to ComfyUI, InvokeAI's node editor is one part of a broader application rather than the entire interface, and it pairs the graph view with a purpose-built canvas for layered editing. Compared to AUTOMATIC1111 and its forks like Forge, InvokeAI leans toward a more polished, opinionated production interface rather than a tab-per-feature extension-driven layout. Compared to Fooocus, which minimizes exposed settings for simplicity, InvokeAI exposes considerably more manual control aimed at users who want repeatable, professional-grade pipelines.
Technical context
InvokeAI’s managed models and canvas retain family-specific pipeline and conditioning boundaries.