Stable Diffusion inside Krita, not another browser tab

What it does

This Krita plugin wires generative AI directly into digital painting workflows. It lets you inpaint selections, expand canvases, upscale images, and generate new content using text prompts or reference images—all without leaving the editor. Under the hood it uses ComfyUI as a backend, but the node graph stays hidden behind an artist-facing interface.

The interesting bit

The project treats diffusion less like a separate app and more like another brush in the toolbox. You can restrict generation to a selection, guide it with scribbles or depth maps via ControlNet, assign different prompts to layer-defined regions, and watch the canvas reinterpret itself in real time with Live Painting. The goal is control without the usual parameter paralysis.

Key highlights

• Supports Flux 2, Stable Diffusion XL/1.5, Illustrious, and Z-Image models for local generation.
• ControlNet integration for scribble, line art, pose, depth, canny edge, and segmentation guidance.
• Live Painting mode reinterprets the canvas in real time as you work.
• Runs locally on NVIDIA, AMD, Intel, and Apple Silicon hardware, or connects to remote ComfyUI servers.
• Job queue and generation history let you iterate without losing previous results or prompts.

Caveats

• Local generation wants at least 6 GB of VRAM; the README warns that less memory means long waits or outright failures.
• AMD and Intel GPU support exists, but the sources note it may require custom ComfyUI setup or come with performance trade-offs.
• It is tightly coupled to Krita 5.2.0 or newer; this is not a standalone image editor.

Verdict

Worth installing if you already work in Krita and want generative AI to feel like a native layer operation rather than a browser detour. Skip it if you need a standalone generator or your hardware sits below the 6 GB VRAM floor.