Image Extender, an MIT-licensed open-source web application created by GitHub user boona13, has gained 237 stars for its approach to AI-powered image outpainting. Released on May 26, 2026, the tool enables users to extend images in any direction while using Poisson blending techniques to create mathematically invisible seams between original and AI-generated content.
Google Gemini Models Power Five Specialized Creation Modes
The application leverages Google's Gemini image models through OpenRouter, with Gemini 3.1 Flash Image as the default option. Users can also select Gemini 3 Pro Image and Gemini 2.5 Flash Image, with each extension operation costing approximately $0.03. The tool offers five operational modes:
- Extender: Standard outpainting with best-of-3 variant picker for comparing results
- Parallax Studio: Multi-layer background creation with Sky/Far/Mid/Near depth layers and live scrolling preview
- Tile Studio: Generates complete 13-tile autotile sets in single AI calls with corner reconciliation
- Sprite Studio: Character animation sheets supporting humanoid, quadruped, serpent, flyer, and blob body plans
- Props Studio: Procedurally generates decoration sprites using art director-to-painter two-call pipeline
Poisson Blending Pipeline Eliminates Visible Seams
The distinctive technical feature is the Poisson blending pipeline, which performs color-drift correction and iterative Gauss-Seidel optimization. This gradient-domain image editing approach represents a more sophisticated solution than the simple alpha compositing used by many AI image generation tools, creating seamless transitions between original and generated content.
Game Engine Integration and Privacy-First Architecture
Built with Next.js 14, React 18, TypeScript, and HTML5 Canvas for client-side processing, Image Extender uses Tailwind CSS and JSZip for bundling exports. The application is privacy-first—API keys remain in browser localStorage and servers never persist credentials. Export integration includes JSON manifests compatible with Unity, Godot, Phaser, and Tiled game engines, positioning it as a practical tool for game developers and digital artists. The interface supports keyboard-native workflows with arrow keys for extensions and dedicated shortcuts for cycling and accepting variants.
Key Takeaways
- Image Extender uses Poisson blending with Gauss-Seidel optimization to create mathematically invisible seams between original and AI-generated content
- The tool offers five specialized modes including Parallax Studio for multi-layer backgrounds and Tile Studio for complete 13-tile autotile sets
- Google Gemini models accessed through OpenRouter power the AI generation at approximately $0.03 per extension operation
- Export formats include JSON manifests compatible with Unity, Godot, Phaser, and Tiled game engines
- The privacy-first architecture stores API keys only in browser localStorage with no server-side credential persistence