Krita AI Diffusion is a plugin that lets you call generative AI from inside the image editor Krita, using nothing but text prompts to inpaint, outpaint, upscale, and refine a selected region. It is released as open source (GPL-3.0) and supports Windows, Linux, and macOS. Under the hood it uses ComfyUI as its backend, and in addition to running on a local GPU, it offers “Interstice,” a cloud generation service for users who do not have a GPU. Development has remained active since the plugin’s release in September 2023; August 2026 brought support for segmentation control layers (v1.53.0), and updates to supported models and usability continue to arrive.
Key Features
- Inpainting and outpainting on selections: Combined with Krita’s layer and selection tools, you can fill, extend, add to, or remove from just the selected region using a text prompt. The result is automatically blended to match its surroundings
- Live painting: A mode in which the AI interprets your canvas in real time as you draw, so you can keep painting while watching the generated result
- Composition control via control inputs: Multiple control layers ─ scribbles, line art, Canny edges, pose, depth, normals, segmentation, and more ─ let you specify composition and poses, producing the image you intended without fine-grained tweaking
- IP-Adapter and regions: Includes IP-Adapter, which transfers style or composition from a reference image (face swapping is supported too), and a regions feature that lets you assign a separate prompt to each layer
- Broad diffusion model support and upscaling: Supports multiple model families including Flux 2, Z-Image, and Stable Diffusion 1.5/XL/Illustrious, and can upscale to 4K, 8K, and beyond. Model support is updated continuously ─ for example, segmentation control for Anima was added in August 2026
Pricing
| Plan | Price | Details |
|---|---|---|
| Plugin itself | Free | Open source (GPL-3.0). No additional cost for running on a local GPU |
| Interstice cloud (one-time, small) | 10€ | 5,000 tokens (no expiration) |
| Interstice cloud (one-time, large) | 28€ | 15,000 tokens (no expiration) |
Pricing is current as of August 2026. Token consumption varies by generation mode and resolution (for example: 1 token for live generation at 640x512, 10 tokens for standard generation at 1024x1024 with SDXL, and 58 tokens at 3840x2160). For the latest pricing, including whether free trial tokens are available, check the official website.
Pros and Cons
✅ Pros
- The plugin itself is open source (GPL-3.0) and free. With a local GPU, you can keep using it at no additional cost
- Because it integrates with Krita’s layer and selection tools, you can use AI generation without leaving your existing drawing workflow
- Multiple control inputs ─ scribbles, line art, depth maps, poses, and more ─ let you specify composition precisely, making it easier to reproduce the image you have in mind
- Users without a GPU can switch to Interstice’s cloud generation service
- With ComfyUI as the backend, it supports a wide range of models, including Flux 2, Z-Image, and Stable Diffusion 1.5/XL/Illustrious
⚠️ Cons
- Running locally is said to require an NVIDIA GPU with at least 6GB of VRAM; on systems that do not meet this, generation becomes very slow
- Some Krita distributions ─ the Windows Store version, the Steam version, and the Linux Flatpak, among others ─ are not recommended, so you need to check the installation method beforehand
- Cloud generation is billed per token, so costs add up as usage grows
Comparison with Similar Services
| Criteria | Krita AI Diffusion | Photoshop Generative Fill | Stable Diffusion WebUI | InvokeAI |
|---|---|---|---|---|
| Provider | Acly (independent developer) | Adobe | AUTOMATIC1111 and the OSS community | Invoke AI |
| Integration | Plugin inside Krita | Built-in Photoshop feature | Standalone web app | Standalone app (custom canvas UI) |
| License | Open source (GPL-3.0) | Commercial (a Photoshop feature) | Open source (AGPL-3.0) | Community edition is open source (Apache-2.0) |
| Runtime | Local GPU or cloud | Cloud only | Mainly local GPU | Local GPU or cloud |
| Pricing model | Free plugin + pay-as-you-go cloud | Included in the Photoshop subscription | Free | Community edition free, higher tiers paid |
Who Is It For?
- People who draw in Krita and want to try generative AI inpainting and outpainting without leaving their existing workflow
- Illustrators who want to produce the images they intend efficiently, specifying composition and poses in detail through control inputs
- People without a GPU who want to try AI generation inside Krita via a cloud generation service
- People who want to run AI image generation locally and for free using open-source tools
Summary
Krita AI Diffusion is an open-source plugin that embeds generative AI inside the image editor Krita, letting you inpaint and outpaint selected regions and specify composition through control inputs without disrupting your existing drawing workflow. It is free to use with a local GPU, and its flexibility in switching to Interstice’s cloud generation service when no GPU is available is a defining feature. With ComfyUI as the backend, it supports multiple diffusion models, and new features have continued to arrive since its release in September 2023.