NanoScene AI

Inspiration

NanoScene AI started from a simple problem: AI image and video creation often gets split across too many tools. A creator may generate an image in one place, edit it somewhere else, upscale it in another workflow, then lose track of the final assets. I wanted to turn that into one focused browser workspace.

For OpenAI Build Week, I used Codex with GPT-5.6 to meaningfully extend and harden NanoScene AI as a production-style AI media product rather than a one-off demo.

What it does

NanoScene AI lets users create AI images from prompts or reference images, generate short AI videos, manage created assets, and continue creative actions from saved outputs.

Key workflows include:

• Prompt-based AI image generation
• Reference-guided image generation and editing
• Image upscaling and background-oriented creative workflows
• Credit-based AI video generation
• An assets library for viewing, downloading, remixing, editing, or turning images into video
• Model-specific public pages for different AI image workflows

How I built it

The app is built as a full-stack TypeScript SaaS using React, TanStack Start, Vite, Nitro, Tailwind CSS, Cloudflare Workers, Cloudflare D1, Cloudflare R2, Drizzle ORM, TanStack Query, Better Auth, Stripe, and provider-backed AI generation APIs.

Codex was used as an engineering partner across the build: reading the live codebase, planning scoped architecture changes, refactoring shared image/video workspace boundaries, improving generation task recovery, unifying model configuration, and tightening the assets workflow.

Challenges

The hardest part was keeping the product coherent while multiple media workflows share infrastructure. Image generation, video generation, asset history, pricing, authentication, storage, and SEO pages all touch different layers of the app. Codex helped keep changes grounded in the actual repository instead of drifting into isolated prototype code.

Another challenge was making the project eligible as an existing product that was meaningfully extended during Build Week. I focused the Build Week work on real user-facing improvements: continuity for active generation tasks, cleaner image/video navigation, asset-to-edit actions, and a more maintainable workspace architecture.

What I learned

Codex works best when it has strong project context: source-of-truth docs, clear architecture boundaries, and narrow implementation scopes. The biggest lesson was that AI coding is not only about generating code faster. It is about keeping product decisions, implementation, validation, and maintenance aligned while the codebase grows.

Built With

• cloudflare
• codex
• css
• d1
• drizzle
• gpt-5.6
• nitro
• openai
• orm
• r2
• react
• start
• tailwind
• tanstack
• typescript
• vite
• workers