Krita AI API Diffusion

A Krita plugin that brings the FLUX models from Black Forest Labs (BFL) and Google's Nano Banana 2 (Gemini image models) directly into Krita — for text-to-image, context-based inpainting and mask-based inpainting, with an optional reference image. The interface is multilingual (English / Deutsch / 中文 / ไทย).

Features

• Two providers side by side — pick a BFL FLUX model or a Google Nano Banana 2 model from the same dropdown; the plugin routes to the right API automatically.
• New image (text → image) — generate an image from a prompt.
• Context Inpaint — edits the current selection context-based (FLUX.2 input_image or Gemini image editing, no mask needed). Great for instructions like “remove the person” or “replace the sky”.
• Mask Inpaint (FLUX.1 Fill) — precise filling exactly inside the selection via image + mask (flux-pro-1.0-fill). FLUX only.
• Reference image (optional) — an additional image used as a template, e.g. a specific pair of sunglasses that a selected cat should wear.
• Batch / variants — generate up to 8 results per click, each with its own seed, collected into an AI Batch layer group so you can pick the best.
• Seed control — random seed by default; lock a seed (or roll a new one) to reproduce a result or make controlled variations.
• Aspect-ratio presets — 1:1, 3:2, 16:9, 4:3, … (or custom width/height).
• Cancel button — abort a running request instead of waiting for the poll.
• Multilingual UI — switch language at runtime under Options.
• Adaptive crop with context padding, feathered mask edges (no visible seams), correct color handling and readable error messages.

Requirements

• Krita 5.x (includes Python/PyQt5 in the official builds).
• For FLUX: a BFL API key from https://api.bfl.ai (the BFL dashboard).
• For Nano Banana 2: a Google AI Studio API key from https://aistudio.google.com/apikey.
• You only need the key(s) for the provider(s) you actually use. Both are paid / metered services.

Installation

Option A — packaged ZIP (easiest)

1. Grab krita-ai-api-diffusion.zip from the repo’s Releases (or build it yourself: run python build_zip.py, which writes it to dist/).
2. In Krita: Tools → Scripts → Import Python Plugin from File and select the ZIP.
3. Open Settings → Configure Krita → Python Plugin Manager, enable Krita AI API Diffusion, then restart Krita.
4. Show the docker via Settings → Dockers → Krita AI API Diffusion.

Option B — manual

1. Download or clone the repository:

   git clone https://github.com/DenRakEiw/krita-ai-api-diffusion.git

2. Copy the ai_api_diffusion folder and the ai_api_diffusion.desktop file into Krita’s pykrita directory:

   OS	Path
   Windows	%APPDATA%\krita\pykrita\
   Linux	~/.local/share/krita/pykrita/
   macOS	~/Library/Application Support/krita/pykrita/

   The result should look like this:

   pykrita/
   ├── ai_api_diffusion.desktop
   └── ai_api_diffusion/
       ├── __init__.py
       └── ai_api_diffusion_docker.py
   

   If the .desktop file is missing, create it with this content:

   [Desktop Entry]
   Type=Service
   ServiceTypes=Krita/PythonPlugin
   X-KDE-Library=ai_api_diffusion
   X-Python-2-Compatible=false
   Name=Krita AI API Diffusion
   Comment=Plugin for Flux (BFL)

3. Start Krita.

4. Open Settings → Configure Krita → Python Plugin Manager, enable Krita AI API Diffusion, then restart Krita.

5. Show the docker via Settings → Dockers → Krita AI API Diffusion.

Option C — clone directly into pykrita

cd <pykrita-directory>
git clone https://github.com/DenRakEiw/krita-ai-api-diffusion.git ai_api_diffusion

Then copy the .desktop file one level up (into pykrita/) as shown above and continue with steps 3–5.

Setup

1. In the docker, switch to the Options tab.
2. Pick your Language (English / Deutsch / 中文 / ไทย) — applied instantly.
3. Enter your API Key (BFL) and click Save. (The key and language are stored in Krita’s settings.)

Usage

New image

1. Generate tab, choose a model (e.g. flux-2-pro).
2. Enter a prompt; optionally set width/height under Options.
3. Click New image (text → image) — the result is added as a new layer.

Context Inpaint (FLUX.2)

1. Select the area to change with a selection tool.
2. Enter an edit instruction as the prompt, e.g. “remove the object”.
3. Click Context Inpaint (FLUX.2).
4. Only the selected area is replaced (softly blended in).

Mask Inpaint (FLUX.1 Fill)

1. Select an area (min. ~256 px — small selections are upscaled).
2. Optionally enter a prompt.
3. Click Mask Inpaint (FLUX.1 Fill) — precise filling inside the mask.

With a reference image (example: sunglasses on a cat)

1. Select the head/eye area of the cat.
2. Click Choose reference image … and load the sunglasses image.
3. Choose model flux-2-pro or flux-2-max (they support multi-image).
4. Prompt: “the cat is wearing these sunglasses”.
5. Click Context Inpaint (FLUX.2).

Models

Selection	Endpoint	Use
flux-2-pro	flux-2-pro	Default, good quality, multi-image
flux-2-max	flux-2-max	Highest quality, multi-image
flux-2-flex	flux-2-flex	Strong at typography/text
flux-2-klein-9b	flux-2-klein-9b	Fast, lightweight (no reference image)
Mask Inpaint	flux-pro-1.0-fill	Mask-based filling

Tips & troubleshooting

• HTTP 422 “Image dimensions must be at least 256x256” — selection too small. It is upscaled automatically; still, select a larger area for sharp results.
• HTTP 402 / “insufficient credits” — no credits on the BFL account.
• HTTP 401 / 403 — API key wrong or not set (Options tab).
• Payload too large / timeout with reference images — lower MAX_EDGE in ai_api_diffusion_docker.py or shrink the reference image beforehand.
• Error messages appear in the status field at the bottom and as a dialog (truncated).

Notes

• BFL result URLs are short-lived; the plugin downloads the images immediately.
• The reference image is only used with FLUX.2 (context inpaint / generation), not with FLUX.1 Fill.

License

Use at your own risk. The FLUX models are subject to Black Forest Labs’ terms of use and pricing.