Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Image & Video Generation

Krita AI Diffusion Review: Open-Source AI Painting Plugin

·7 min readBy Hans Kuepper · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

Krita AI Diffusion (version 1.53.0) is a GPL-3.0 plugin that brings selection-based generative fill, live painting and upscaling into Krita, and it can run on your own graphics card or on the paid Interstice.cloud service. It generates images with diffusion models rather than text with a language model, and it needs Krita 5.2 or newer plus a model download for local use.

Key Takeaways

  • Maintenance: the latest release, 1.53.0, dates from 22 August 2026 and the main branch changed on 3 October 2026, so no stale-project warning applies.
  • Scope: it generates images with diffusion models; it is not a language-model (LLM) app, and the PromptQuorum directory lists it among its image tools.
  • Cost: the plugin and local use are free; the online service sells prepaid token packs with no subscription.
  • Platforms: Windows, Linux and macOS, with Krita 5.2 or newer.
  • Operator: the service is run by Cosmic Interstice UG in Mannheim, Germany; the plugin code is open under the GPL-3.0 license.

📍 In One Sentence

Krita AI Diffusion is an open-source plugin that adds generative fill, live painting and upscaling with diffusion models to Krita, running on your own GPU or on the paid Interstice.cloud service.

💬 In Plain Terms

Think of it as an AI brush inside a normal painting program: you select an area, describe what should appear there, and the generated pixels land on a layer you can keep editing.

📌Note: This review is based on interstice.cloud, its documentation, privacy policy and imprint, and the GitHub repository, checked on 30 October 2026. PromptQuorum has not installed the plugin or generated images with it, so image quality and speed are not assessed.

Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

Krita AI Diffusion is an open-source plugin that adds generative fill, live painting and upscaling to the free painting program Krita. It runs diffusion image models, not language models, either on your own graphics card or on the paid online service at Interstice.cloud. This review is based on the Interstice website, its documentation and privacy policy, and the GitHub repository, all checked on 30 October 2026; PromptQuorum has not tested the plugin hands-on.

Krita AI Diffusion Review: Open-Source AI Painting Plugin

What Is Krita AI Diffusion?

Krita AI Diffusion is a plugin for Krita, the free painting program, that sends selections and sketches to a diffusion model and returns the result as a layer. The README says its goals are precision and control, integration into normal editing workflows, and local, open, free operation, with cloud generation offered as a way to start without heavy investment.

The project and the service share one name on the web, and the sources describe them separately. The GitHub repository belongs to the account Acly and lists interstice.cloud as its homepage, while the imprint names Cosmic Interstice UG as operator of Interstice.cloud; the sources do not spell out the legal relationship between the two. The plugin uses ComfyUI as its backend.

Get It

Krita AI Diffusion is a ZIP file that you import into Krita; the local server and its models are installed afterwards from the plugin's Configure dialog.

Option
How to get it
PluginZIP from the GitHub release page
KritaVersion 5.2 or newer from krita.org
Installation guidedocs.interstice.cloud/installation
Websiteinterstice.cloud
Privacy policyinterstice.cloud/privacy
Source codegithub.com/Acly/krita-ai-diffusion, GPL-3.0 license

Version reviewed: 1.53.0, published on 22 August 2026 per the GitHub Releases page, which marks it as the build for Krita 5; the newest change to the main branch was on 3 October 2026. No terminal is mentioned in the installation steps. The guide lists a 2.x plugin line for Krita 6 as coming soon. In Krita, import the ZIP via Tools > Scripts > Import Python Plugin from File; the guide advises the krita.org build of Krita, not the Windows Store or Steam version.

Requirements

Local generation needs a capable graphics card: the README recommends an NVIDIA card with at least 6 GB of VRAM, and the guide names an RTX 2060 or better. The CPU also works, but both documents call it very slow.

Operating system

What the sources say:
Windows, Linux and macOS; on a Mac only Apple silicon (M1 to M5) is listed, and the README names macOS 14 or newer

Graphics hardware

What the sources say:
NVIDIA via CUDA, AMD via ROCm, Intel Arc with 8 GB or more via XPU, Apple silicon via MPS; the README lists AMD and Intel for Windows and Linux

Disk space

What the sources say:
At least 10 GB free for the managed server; models can take 50 GB or more depending on the optional packages chosen

Models

What the sources say:
Chosen as workloads during the server install, or added later as individual packages; each workload downloads its recommended models

Own ComfyUI server

What the sources say:
Optional: an existing local or remote ComfyUI server works if it has the required custom nodes and models listed in the ComfyUI setup guide

Internet

What the sources say:
Needed for the downloads and for the online service; the privacy policy mentions offline use, but the installation guide does not describe it

RAM requirements are not stated in the sources. Hardware lists are the maker's; PromptQuorum did not test which cards work in practice.

Features Confirmed by the Sources

Every item below comes from the README or the website; none has been independently tested.

  • Inpainting and live painting. Selections drive generative fill, expand, add and remove, and Live Painting lets the AI interpret the canvas in real time.
  • Upscaling. The README says images can be upscaled to 4k, 8k and beyond without running out of memory.
  • Models. The README lists Flux 2, Z-Image, Stable Diffusion 1.5, Stable Diffusion XL and Illustrious, plus edit models that change an image from a text instruction; the online service page names Stable Diffusion 1.5 and XL.
  • Control. ControlNet modes such as scribble, line art, Canny edge, pose, depth, normals and segmentation, IP-Adapter reference and style images, and regions that give each layer-defined area its own text.
  • Workflow. A job queue, a browsable history of results and prompts, and custom presets with your own checkpoints, LoRA files and samplers.

Plans and Cost

The plugin and local generation are free, and the optional online service sells prepaid token packs instead of a subscription. The service prices below are in euros as the website states them.

Local

What the sources list:
Free: the plugin under the GPL-3.0 license, run on your own hardware with a managed or custom server; the website calls it "100% free on local hardware"

Online service

What the sources list:
Prepaid tokens that never expire: 5,000 for €10 or 15,000 for €28, plus some free tokens to try; a 1024 x 1024 image costs about 5 tokens (about €0.01) and a live-mode image about 1 token

Figures as shown on the service page on 30 October 2026; actual token cost depends on resolution, strength and workflow and is shown in the plugin, and taxes were not checked.

Data Flow and Privacy

The privacy policy, last updated on 4 April 2024, says offline use of the plugin involves no information exchange or collection, and it also says plugin use generates performance metrics that are not linked to personal information. The sources do not reconcile the two statements or say whether that telemetry applies to local use.

For the online service, the website says input images are used only to generate results and results are not stored, although temporary storage may be used for large images. The policy adds that personal data is stored in Germany, that data needed to generate images may be processed in other European countries, Canada or the US, that images and text prompts are excluded from service-improvement analysis, and that it does not state how long generated images are kept. Exports from the service carry an AI-generated metadata tag, and customers in the EU also get an invisible watermark.

📌Note: These are the maker's declarations, not audit results. The plugin code is public under the GPL-3.0 license, but PromptQuorum has not read it or inspected network traffic, and the policy is older than the current release.

What We Could Not Verify

  • Image quality and speed. PromptQuorum did not generate images, time any hardware or compare results with other tools.
  • Model licenses. The GPL-3.0 license covers the plugin code; each diffusion model has its own terms, including for commercial use, and we did not check any of them.
  • Hardware in practice. The VRAM and card lists are the maker's; which cards run Flux 2 or the larger models comfortably is not stated.
  • Telemetry in local use. Whether the performance metrics mentioned in the policy are sent when running locally is not explained.
  • Legal link between project and service. The sources do not describe how the plugin author and Cosmic Interstice UG are related.

Competitors and Alternatives

ComfyUI

Platforms:
macOS, Windows, Linux
Price / license:
Free / GPL 3.0
Key difference:
A node-graph editor you operate directly; this plugin uses ComfyUI as its backend but hides it behind Krita
Articles about ComfyUI (10)

+14 more not shown

Invoke AI

Platforms:
macOS, Windows, Linux
Price / license:
Freemium / Apache 2.0
Key difference:
A separate web-based app with its own canvas rather than a plugin inside a painting program
Articles about Invoke AI (1)

Stable Diffusion WebUI Forge

Platforms:
macOS, Windows, Linux
Price / license:
Free / AGPL-3.0
Key difference:
A parameter-heavy local web UI for generation, with no Krita integration
Articles about Stable Diffusion WebUI Forge (2)

Also mentioned in:

Draw Things

Platforms:
macOS, iOS
Price / license:
Freemium / closed source
Key difference:
A native Apple app with its own canvas and inpainting, not tied to Krita
Articles about Draw Things (2)

Also mentioned in:

These are same-category apps from the PromptQuorum directory; confirm each app's current price, license and platforms on its own page.

Compare in directory

Frequently Asked Questions

Is Krita AI Diffusion the same as Interstice.cloud?

Not exactly: the plugin is the open-source part, while Interstice.cloud is the website and paid online service the plugin can connect to. The plugin's Configure dialog lets you pick the online service, a managed local server or your own ComfyUI server.

Does it work with Krita 6?

Not yet according to the installation guide: Krita 5.x uses the 1.x plugin line, and a 2.x line for Krita 6 is listed as coming soon. The 1.53.0 release notes recommend Krita 5.3.2.1.

Can I use my own models?

The README says you can customize presets with custom checkpoints, LoRA files and samplers, and the online service page mentions uploading your own LoRA models. We did not test either.

Do I need a GPU?

For comfortable local use, yes; the CPU works but is described as very slow. The online service needs no GPU of your own.

Is the output safe to use commercially?

The service page says users own the images they generate, but each diffusion model carries its own license for local use and the sources do not summarize them, so check the license of the model you pick.

Verdict

Krita AI Diffusion is an actively maintained, open-source way to add generative fill and live painting to Krita, with a local option on your own GPU and a prepaid cloud option for machines without one. It suits painters who already work in Krita and accept a plugin install plus multi-gigabyte model downloads. The privacy statements are the maker's own and partly unreconciled, the policy predates the current release, Krita 6 is not yet supported, and image quality, speed and model licensing are untested here, so try it on a small selection and read the license of your chosen model before commercial use.

Sources

← Back to Power Local LLM

MCP