Prompt Enhancement Engine

What It Does

You give it a creative brief and reference images. It gives you back an optimised art direction brief — the kind an experienced art director would write — plus a full set of structured prompts for image generation, image editing, and video creation tools.

Prompts are generated with consistency across them, so you can iterate on a visual concept across multiple tools without losing coherence. The goal is to dramatically compress the time between a creative idea and high-quality AI output.

Who It's For

AI artists and creative teams who are iterating on generative visual projects. Also useful for anyone who isn't deeply fluent in prompt writing but needs reliable, high-quality AI outputs.

Why I Built It

Primarily to speed up my own workflow. Prompt engineering is genuinely one of the highest-leverage skills in AI visual production right now, but it's also slow and repetitive when done manually across multiple tools. Automating the structured part — brief interpretation, prompt formatting, consistency between shots — frees up attention for the creative decisions that actually matter.

It also became a useful proving ground for agentic engineering patterns: streaming responses, multi-model orchestration, structured JSON validation from LLM output, and keeping a UI fast and lean without reaching for a component library.

Technical Stack

• Framework: Next.js (App Router, server routes, Turbopack) · React 19 · TypeScript
• Styling: Tailwind CSS v4 — all UI hand-rolled, no component library
• LLM / API: OpenRouter — raw fetch, no SDK. Primary models: Gemini 2.5 Flash (vision) and DeepSeek v3.2 (briefs/prompts). Fallbacks: Gemma 4 31B and Qwen 3
• Validation: Zod — runtime validation of all LLM JSON responses
• Deployment: Vercel — serverless, SSE streaming via ReadableStream
• Notable: No database, no auth, no state management library. Zero dependencies beyond React. All state is useState/useRef.