Invoke is a self-hosted image generation application for artists and creative professionals. It runs a local web server with a React interface, and you generate, refine, and iterate on images on your own hardware rather than through a cloud service.
Work happens in two main places. The Unified Canvas is a layer-based editing surface with in-painting, out-painting, and brush tools, used to extend AI output, sketches, photography, or renders. Node-based workflows expose the same generation steps as a visual graph, so you can build reproducible workflows, surface selected parameters in the interface, and share workflows with others.
A model manager handles checkpoints, diffusers models, some GGUF models, LoRAs, textual inversion embeddings, and ControlNets. Supported model families include Stable Diffusion 1.5, 2.0, SDXL and 3.5, CogView 4, the Flux.1 and Flux.2 lines, Z-Image, Krea 2, Qwen Image, and ERNIE-Image, with Nano Banana, GPT Image, and Wan available through their APIs only.
Features
- Unified Canvas: layer-based canvas with in-painting, out-painting, brush tools, and every core generation capability
- Node workflows: visual graph editor for building, saving, and sharing reproducible generation workflows
- Board and gallery: organizes generated images into boards, with embedded metadata to recall the prompt and settings behind each image
- Model manager: manages checkpoints, diffusers and some GGUF models, LoRAs, embeddings, and ControlNets
- ControlNet guidance: steers composition with depth maps, edges, and poses
- Iteration tools: batch generation, prompt wildcards, and high-resolution upscaling inside the app
- Segmentation: object segmentation and selection with SAM and SAM2 models
- Local web UI: runs on your own hardware as a local web server with a React interface
