Stable Diffusion's Missing Studio: Canvas, Nodes, and Local Control

What it does

Invoke is a locally hosted creative engine for Stable Diffusion and a growing list of other image-generation models. It wraps inference in a React web interface that combines a full painting canvas, node-based workflow pipelines, and gallery management. The goal is to let professionals iterate on and refine generated imagery without treating the model as a simple text-in-image-out vending machine.

The interesting bit

The Unified Canvas blurs the line between generation and editing: you can inpaint, outpaint, and brush over existing images while the AI fills in the gaps. Meanwhile, the node-based workflow system lets you build reusable, shareable generation pipelines without abandoning the UI. It is the rare local tool that tries to replicate a production art pipeline instead of just running inference.

Key highlights

• Runs a local web server with a React UI; no cloud dependency for core features
• Supports a wide model zoo including SD 1.5/2.0/XL/3.5, Flux variants, CogView 4, and Qwen Image, plus segmentation via SAM/SAM2
• Unified Canvas supports in/outpainting, brush tools, and augmenting existing sketches or photography
• Node-based workflows for customizable generation pipelines that can be shared
• Drag-and-drop gallery with metadata recall for prompts and settings

Caveats

• Several listed models (Nano Banana, GPT Image, Wan) are API-only, so the tool is not fully offline for the entire model catalog
• Hardware requirements are vague; the README only mentions “compatible hardware” without specifics

Verdict

Worth a look if you want a local, browser-based studio for diffusion models with an emphasis on canvas editing and reusable workflows. Less compelling if you were hoping for a lightweight headless tool or full local support for every listed model.