Switch

About the Project

Switch was inspired during a late–night conversation with a startup founder where we discussed the limitations people face when trying to experiment with generative AI casually. That same night, I attempted to change the clothes in one of my photos using an LLM (specifically a model similar to Google’s Nano Banana). Instead of only changing the outfit, the model altered my entire posture, face, and background — something I didn’t want at all.

This highlighted a real problem:
Most GenAI tools are not built for simple, fun, everyday fashion experimentation.
Users must deal with complex prompt engineering, and models often fail to preserve the original person in the image.

That’s when I realized that many people — not just designers, but casual users — face the same issue. So I set out to build a platform that solves it.

What I Built

Switch is a modern, AI-powered platform that lets users:

Generate new clothing styles using simple, friendly prompts
Try those clothes on any uploaded image, while preserving posture, identity, and background
Apply art-style modifiers (vintage, anime, cyberpunk, minimal, etc.)
Skip complex prompt engineering thanks to backend prompt-sanitization for optimal output

The goal is to make fashion experimentation fun, reliable, and accessible — whether you’re a designer brainstorming ideas or someone casually trying out new looks.

How I Built It

Switch is built using:

React + Vite for a fast, smooth front-end
TailwindCSS for a modern dark-theme interface (black + orange accent)
AI image generation backend based on a fine-tuned fashion try-on model
Prompt sanitization engine that transforms user prompts into structured, model-friendly instructions
Dynamic art-style pipelines built using model modifiers

This ensures:

Identity preservation
High-quality outfit generation
Support for style transfer

What I Learned

Building Switch taught me:

How to work with image-to-image AI models and tune them for identity consistency
The importance of prompt cleaning/sanitization to get reliable outputs
How to design a user-friendly fashion interface that hides technical complexity
The challenges of combining multiple AI pipelines
How to create a clean UX for a creative tool

Challenges I Faced

Maintaining identity and posture during outfit replacement
Handling inconsistent model outputs and stabilizing them with prompt rules
Designing a UI that feels professional yet playful
Optimizing performance for low-latency, high-quality generation
Supporting different art styles consistently

Conclusion

Switch transforms the way people interact with fashion and AI.
It removes the barriers of prompt engineering, preserves user identity, and makes outfit generation effortless. What began as a personal frustration evolved into a full platform that helps anyone experiment with clothing and art styles — powered by clean engineering, thoughtful UX, and modern AI.