Nanobanana AI Chat

Nanobanana AI Chat is a specialized Krita plugin acting as an AI Copilot for digital artists. It integrates Google's Gemini 2.5 Flash Image model directly into Krita, allowing for conversational image generation, editing, and rapid variation creation.

Features

• Copilot Interface: Compact UI designed to sit in a dock without taking up screen space.
• Chat Session (History): The model remembers context (e.g., "Make it blue" works because it remembers the previous image).
• Batch Generation: Generate 1 to 4 variations in a single click.
  • Variation 1: Context-aware (updates chat history).
  • Variations 2-4: Stateless (parallel-like exploration without polluting history).
• Img2Img Support: Sends the current canvas visualization to the AI as reference.
• Real-time Streaming: Text appears instantly; images load as layers as soon as they are ready.

Architecture

The project uses a thread-safe, asynchronous architecture to ensure Krita's UI remains responsive during AI generation.

• nanobanana.py: The main UI controller using PyQt5.
• gemini_api.py: Handles communication with Google GenAI (Gemini 2.5/3.0).
• krita_api.py: Bridge between Krita and Python.
• utils.py: Manages persistent settings.

Technology Stack

• Python 3.10+ (Embedded in Krita)
• PyQt5: UI and Threading.
• Google GenAI SDK (google-genai): Interaction with Gemini models.
• Pillow (PIL): Image processing.

Setup

1. Install: Place nanobanana folder in Krita's pykrita directory.
2. Configure: Open Settings (⚙️) in the docker and paste your Google AI Studio API Key.
3. Usage:
   • Select Aspect Ratio.
   • Set Batch Count (x1 - x4).
   • Check Img2Img if you want to edit the current drawing.
   • Type prompt and hit Generate.

Development

• Tests: Run pytest tests/ -v to verify API logic and Utils.