How We Built a "Full-Stack AI App" in Just 8 Days
How We Built a "Full-Stack AI App" in Just 8 Days
By Tomasz Cholewa, June 2025 In an ambitious eight-day sprint, a "full-stack app" was developed using AI, achieving a 100% success rate. This project centered around creating an application that could transform a user's simple text prompt into a visually impressive image featuring the MGS5 EV.
The Challenge
The project, which commenced on a Friday, began with a Figma wireframe and a straightforward brief. The core objective was to build an app capable of generating a "wow" image from a user's prompt. However, this came with a set of significant constraints:
- No Dedicated Infrastructure: There was no time or budget to build and maintain dedicated infrastructure. * Performance Target: The image generation process had to be swift, delivering the final result in under 30 seconds from the moment the user entered the prompt. * Accuracy Requirement: The vehicle in the generated image needed to be a high-fidelity and faithful representation of the actual car model. Initial attempts to meet these requirements were unsuccessful.
The Spark of Inspiration
A breakthrough occurred over the weekend with the implementation of Black Forest Labs' FLUX.1. This instantly resolved most of the challenges. However, it introduced a new issue: the generated vehicles were too repetitive as they were heavily based on the reference image. The solution was to use the FLUX.1 dev model, which enabled quick fine-tuning on a curated set of images to achieve the necessary variety.
From Proof of Concept to "Magic Dust"
By Tuesday, a proof of concept was in the works, but with a twist. A key consideration was the user experience, particularly for those using a touchscreen who were unlikely to type lengthy prompts. The main problem to solve became how to generate a "wow" image from a very simple input. The solution, conceptualized on Wednesday, was to use another AI to "enrich" the user's prompt. This "Magic Dust" approach was the perfect way to turn simple input into a stunning result while also incorporating necessary guardrails.
Application Architecture
The application's architecture, finalized between Thursday and Friday, consisted of several key components:
- Frontend: A Vue.js application running in the user's browser provided the user interface. * Backend: A Python-based Flask API orchestrated the image generation process. It managed prompt enhancement, calls to external services, file storage, and data logging. * External Services: Third-party APIs provided core functionalities. This included OpenAI for AI-powered prompt enhancement and Replicate for image generation. Mailgun was also integrated for email sending capabilities. The system was designed to have the Flask backend receive and enhance a prompt, which was then sent to the Replicate API for image generation, with the final image and other data being stored on a file system.
The Grand Finale
Following a weekend of final polishing, the project was launched on Monday. The application proved to be a resounding success, with over 200 images generated at the event. The system demonstrated its ability to take simple user inputs and, through AI-powered prompt enrichment, create detailed and stylized images. For example:
- An input of "james bond" was transformed into a detailed prompt for a "striking portrait of James Bond standing confidently next to a sleek MGS5 suv in a sophisticated cityscape at dusk." * The Polish phrase for "children's room," "pokój dzieciecy," resulted in an image of a vibrant room featuring a small MG logo-themed bookshelf and a bright MGS5 SUV toy car. * The input "plaża we władysławowie, parawany, plażowicze" (beach in Władysławowo, beach screens, beachgoers) was given a "[cosmic style]" to create a surreal, otherworldly landscape with a majestic MGS5 EV. * A surreal image in the style of Salvador Dalí, featuring melting clocks, was generated from the prompt "obraz czas w stylu dalhi" (image of time in Dali style). * A Halloween scene with eerie, illuminated skeletons and an MGS5 SUV was created from the input "hallowen straszne kosciotrupy" (halloween scary skeletons). This project showcases the potential for rapidly developing innovative applications by leveraging modern AI tools.