Node Banana: Visual Node-Based Workflows for AI Image Generation

By Prahlad Menon 3 min read

If you’ve used ComfyUI, you know the power of node-based workflows for image generation. But ComfyUI is complex, tightly coupled to Stable Diffusion, and has a learning curve that intimidates newcomers.

Node Banana takes a different approach: a clean, web-based node editor focused on simplicity and multi-provider support. Built mainly with Claude Opus 4.5 (yes, really), it’s designed for anyone who wants to chain AI image operations without writing code.

What Makes It Different

Prompt-to-Workflow generation. Describe what you want in natural language, and Node Banana generates the complete workflow for you. “Create a pipeline that takes a sketch, generates a detailed image, then upscales it” becomes a connected node graph automatically. Currently Gemini-powered, with more providers coming.

Multi-provider support. Unlike tools locked to one model ecosystem, Node Banana connects to:

  • Google Gemini (text and image generation)
  • Replicate (thousands of open-source models)
  • fal.ai (fast inference for image/video)
  • OpenAI (text generation)

This means you can mix models in a single workflow. Use Gemini for ideation, Replicate for a specific LoRA, and fal.ai for fast upscaling — all connected visually.

Built-in image annotation. A full-screen editor with drawing tools (rectangles, circles, arrows, freehand, text) lets you annotate images directly in the workflow. Perfect for inpainting masks or visual feedback loops.

How It Works

The interface is straightforward:

  1. Add nodes from the floating action bar — image generators, text processors, annotation tools
  2. Connect outputs to inputs by dragging between handles (type-matched: image→image, text→text)
  3. Configure each node — select models, adjust parameters, set prompts
  4. Run the workflow — execution flows through the graph, each node processing in sequence

Workflows save as JSON files, so you can version control them, share them, or build a library of reusable pipelines.

The Tech Stack

For developers interested in extending it:

  • Next.js 16 with App Router
  • React Flow (@xyflow/react) for the node editor
  • Konva.js for canvas annotation
  • Zustand for state management
  • TypeScript throughout

The codebase is clean and well-organized — a testament to AI-assisted development done right.

Who Should Use This

Node Banana fills a specific gap: you want visual AI workflows without ComfyUI’s complexity, and you want to use multiple AI providers without switching tools.

It’s particularly useful for:

  • Designers building repeatable image generation pipelines
  • Developers prototyping multi-step AI processes
  • Content creators chaining generation → editing → upscaling workflows
  • Anyone who thinks better visually than in code

The project is early (the README warns about rough edges), but the foundation is solid. Chrome recommended for now; join their Discord for support.

Getting Started

git clone https://github.com/shrimbly/node-banana
cd node-banana
npm install
# Add API keys to .env.local
npm run dev

Then open localhost:3000 and start connecting nodes.

What I like most: it doesn’t try to do everything. It’s a focused tool for visual AI workflows, built by someone who actually uses it for their own projects. That pragmatism shows in the design.

Explore Node Banana on GitHub