G/GenAI Studio

GenAI Project

GenAI Studio — Text & Image Generation Lab

Explore. Generate. Experiment. Learn.

Project Overview

GenAI Studio explores practical applications of Generative AI through text generation, image generation and prompt engineering — built as a working product, not a static demo.

Problem Statement

Generative AI can produce very different results depending on how instructions are written. This project explores how structured prompting improves the quality and consistency of AI-generated content.

Objectives

  • Explore Generative AI
  • Understand prompt engineering
  • Build a practical AI-powered application
  • Experiment with text generation
  • Experiment with image generation
  • Compare different prompting approaches
  • Document the development process

Features

  • Dashboard with usage stats and recent work
  • Text generator with type, tone, length and audience controls
  • Image generator with style, mood, lighting and aspect ratio
  • Prompt Engineering Lab with explained rewrites
  • Interactive text vs image prompt builder
  • Searchable prompt library (20+ prompts)
  • Generation history with favourites and reopen
  • Case study documenting experiments

Technologies

  • React 19 + TypeScript
  • TanStack Start (routing & server functions)
  • Tailwind CSS
  • Lovable AI Gateway (text & image models, key kept server-side)
  • Lovable (build platform)
  • Git/GitHub — [add repo link]

Challenges

  • Designing effective prompts that generalise across inputs
  • Managing and parsing AI-generated responses reliably
  • Handling API integration, errors and rate limits
  • Designing a UX that teaches rather than just outputs
  • Maintaining consistent output quality

Lessons Learned

  • Specificity: vague prompts yield generic output
  • Context: models can't infer what they aren't told
  • Constraints: length and tone limits improve focus
  • Iteration: the first prompt is rarely the best
  • Text vs image: briefs vs visual shot lists
  • Limitations: models can be confidently wrong
  • Testing: outputs must be reviewed, not trusted blindly