Stable Diffusion Image Editors in 2026: Which Ones Are Still Maintained

Quick answer

• 🔴 The two interfaces most tutorials still recommend have stopped. AUTOMATIC1111 last released v1.10.1 in February 2025 and has made zero commits in 90 days. Forge last released in February 2024 and has not been pushed to since July 2025.
• If you want an actual canvas — brush, selection, layers — there are two serious local options: InvokeAI (Apache-2.0, released three days before this was written) and the Krita AI Diffusion plugin (GPL-3.0).
• ComfyUI is not an image editor, whatever its popularity. It is a node graph with a mask tool attached.
• Licence matters if this is work: A1111 and Forge are AGPL-3.0. InvokeAI is Apache-2.0 and SwarmUI is MIT.
• Budget 6 GB of VRAM as the practical floor, which is what the Krita plugin recommends.

Repository figures below were read from the GitHub API on 29 September 2026. They date faster than anything else in this guide — re-check before acting on them.

Which editors are actually maintained

This is the question worth asking first, because the answer has changed and most guides have not caught up.

Read the first and last rows together. AUTOMATIC1111 has by far the most stars of anything here — 165,146, roughly six times InvokeAI — and has shipped nothing for over nineteen months. Star counts measure accumulated attention, not current health, and this is the clearest example of that gap you will find in local AI. Forge, the fork people moved to for speed, is further behind still: no release since February 2024, no push since July 2025. The same pattern runs through the AI upscale models these editors load: every major one is dormant while the loaders around them still ship.

Neither is archived and neither is broken — existing installs keep working. But they will not gain support for new models, and that is exactly what dates an image editor. The same has happened to the same author’s other projects, ControlNet among them — we trace that in which ControlNet models still ship.

What each one actually gives you

“Image editor” means different things across these tools, and the distinction matters more than the feature lists suggest.

InvokeAI — the closest to a real editor

Its README describes the Unified Canvas as “a fully integrated canvas implementation with support for all core generation capabilities, in/out-painting, brush tools, and more”. That is the shape most people mean by an editor: an infinite canvas you paint on, mask directly, and extend outward. It is Apache-2.0, which is the most permissive licence of anything in this comparison, and it released three days before this guide was written.

Krita AI Diffusion — generation inside a real paint program

The other honest answer, and a different bet: rather than building an editor around a model, it puts the model inside Krita, a mature open-source painting application. The plugin’s own description is “Streamlined interface for generating images with AI in Krita. Inpaint and outpaint with optional text prompt, no tweaking required”, and its stated aim is that you “Draw, paint, edit and generate seamlessly without worrying about resolution and technical details”.

Two features are worth calling out because nothing else here has them. Inpainting driven by selections — “Use selections for generative fill, expand, to add or remove objects” — and Live Painting, which the project describes as letting the AI “interpret your canvas in real time for immediate feedback”. If you already paint, this is the one to try first. It needs Krita 5.2.0 or newer, and installs through Tools ▸ Scripts ▸ Import Python Plugin from File.

ComfyUI — powerful, but not an editor

ComfyUI has a mask editor and can absolutely inpaint, but the working surface is a node graph, not a canvas. If your task is “fix this one area of this photograph”, you will spend longer wiring than editing. It is the right tool when the workflow is the thing you are building — and it ships 660 templates to start from, which we went through in the ComfyUI templates guide.

SwarmUI — a front end over ComfyUI

MIT-licensed and actively developed, SwarmUI describes itself as “A Modular Stable Diffusion Web-User-Interface, with an emphasis on making powertools easily accessible”. It gives you a conventional tabbed interface backed by ComfyUI underneath. Useful if you want ComfyUI’s capability without its graph, but it is a generation interface first; editing is not its centre of gravity.

The licence difference is not academic

If you are editing images for a client or inside a product, the interface licence matters as much as the model licence, and they are separate questions.

• InvokeAI — Apache-2.0. Permissive. No copyleft obligation.
• SwarmUI — MIT. Permissive.
• Krita AI Diffusion — GPL-3.0, and ComfyUI likewise. Copyleft, but you are running it as a tool, not distributing modified copies, which is the usual case.
• AUTOMATIC1111 and Forge — AGPL-3.0. The most restrictive here. AGPL extends copyleft obligations to network use, which is the clause that matters if you ever put a modified version behind a web service.

And keep this separate from the weights. A permissively licensed editor running FLUX.1 [dev] does not make the model commercially usable — that is a different chain entirely, which we untangle in the Flux licence guide. Our model licence checker covers the weights side.

What hardware you need

The Krita plugin is the only project here that publishes a clear floor, and it is a reasonable one for all of them: “To run locally a powerful graphics card with at least 6 GB VRAM (NVIDIA) is recommended. Otherwise generating images will take very long or may fail due to insufficient memory!”

It lists NVIDIA via CUDA, AMD via ROCm, Intel via XPU, Apple Silicon via MPS on macOS 14+, and CPU as “supported, but very slow”. That is a broader hardware range than most of this software offers. For what a specific model and resolution will actually consume, our image generation VRAM calculator computes peak usage, and the guide to running Stable Diffusion locally covers the tiers.

How to choose

• You want to edit images, with a canvas: InvokeAI. Apache-2.0, active, and the Unified Canvas is purpose-built for it.
• You already use Krita, or you paint: the Krita AI Diffusion plugin. Live Painting is genuinely different from everything else here.
• You are building a repeatable pipeline: ComfyUI, and accept that it is a graph rather than an editor.
• You are following a 2024 tutorial that says install AUTOMATIC1111: check the dates first. It still runs, but it has not shipped a release since February 2025 and will not support newer models.

Whichever interface you pick, the model doing the editing is a separate decision — SD 1.5 Inpainting, SDXL Inpainting, FLUX.1 Fill and Kontext all behave differently and carry different licences. We compare those in the Stable Diffusion inpainting guide.

FAQ

What is the best Stable Diffusion image editor?

InvokeAI, for most people who want an actual editing canvas: it is Apache-2.0, actively released, and its Unified Canvas supports inpainting, outpainting and brush tools in one surface. If you already paint in Krita, the Krita AI Diffusion plugin is the better fit.

Is AUTOMATIC1111 dead?

Not archived, and existing installs still work — but its last release was v1.10.1 on 9 February 2025 and it has made no commits in the last 90 days. Treat it as dormant rather than current, and do not expect support for new models.

Can ComfyUI edit images?

Yes, it has a mask editor and inpainting workflows. But it is a node-graph tool, not a canvas, so for one-off edits to a single image it is slower to use than InvokeAI or Krita.

How much VRAM do I need to edit images with Stable Diffusion?

At least 6 GB is the recommendation published by the Krita AI Diffusion project, and it is a fair floor across these tools. Less will work but generation may be very slow or fail outright.

Sources

• Licences, star counts, release dates and 90-day commit counts read from the GitHub API on 29 September 2026: InvokeAI, Krita AI Diffusion, SwarmUI, AUTOMATIC1111, Forge
• Unified Canvas description, Live Painting, selection-based inpainting and the 6 GB VRAM recommendation: the projects’ own README files

Updated 29 September 2026: all repository figures read on this date. Maintenance status changes — re-check the release dates before relying on them.