Full-Stack AI-Assisted Tattoo Design Platform

Overview

A UK creative studio needed a production web platform to bring AI-assisted design into the tattoo workflow — not as a gimmick, but as a working tool that studios, artists and clients could collaborate through. We took the product from inception to production with total ownership of the full-stack web application, integrating with an existing Python AI backend.

The core loop: a client describes or sketches an idea, the artist generates and refines concepts with AI assistance on a zoomable canvas, iterations are tracked per project, and the final design exports as a print-ready professional file.

The Challenge

• Design iteration was slow. Concept revisions moved over email and chat, with no shared record of what had been tried.
• AI generation needed to feel like a tool, not a toy. Artists needed fine control — masked inpainting on specific regions, style variations and layered output — not one-shot image generation.
• Multiple audiences, one system. Studios, artists, clients and admins each needed different permissions and views over the same projects.
• Assets are the product. High-resolution design files needed secure storage, controlled sharing and reliable delivery.

What We Built

Full-stack Next.js application

A Next.js 14 / TypeScript application with Redux Toolkit and Redux Saga managing the heavier client-side flows — generation queues, canvas state and optimistic updates. Data modelling with Prisma against PostgreSQL, JWT auth enforced in Next.js middleware, and role-based access control across four user types: studio, artist, client and admin.

Zoomable design canvas

The centrepiece is a canvas where artists work at fine detail: zoom into a region, mask it, prompt the AI to regenerate just that area, and overlay results as layers. Finished work exports as layered PSD files, so the platform slots into a professional print workflow rather than replacing it.

AI generation pipeline

Generation runs on a separate Python/Django service orchestrating ComfyUI workflows with HuggingFace and OpenAI models. The web app communicates through a clean API contract — prompt-based generation, region inpainting and style variations — keeping GPU workloads isolated from the user-facing application.

Cloud infrastructure

• AWS S3 with pre-signed URLs for secure upload and delivery of design assets
• AWS RDS (PostgreSQL) for relational data
• AWS SES for transactional email
• Vercel hosting with preview deployments on every pull request

• Next.js 14
• TypeScript
• Redux Toolkit + Saga
• Prisma
• PostgreSQL
• AWS S3
• ComfyUI
• OpenAI

Outcome

• Delivered the full web platform from first commit to production, as the sole engineer on the web stack.
• AI-assisted generation, masked inpainting and PSD export shipped as working studio tools — used in real client sessions, not demos.
• Role-based project spaces replaced email threads as the system of record for design iterations.
• Clean separation between the web app and the GPU generation service means either side can evolve independently — new models slot in behind the same API contract.