What it does Photoshot is a full-stack web app that lets users upload photos, train a personalized DreamBooth model via Replicate, and generate stylized avatars. It wraps the ML pipeline in a Next.js interface with Stripe billing, S3 storage, and a prompt wizard powered by OpenAI.

The interesting bit The project is essentially a well-wired integration layer — it doesn’t train models itself, but it solves the unglamorous problem of turning Replicate’s async API into a paid SaaS product. The “prompt wizard” feature, which uses OpenAI to generate style prompts, is a small but clever UX shortcut that keeps users from staring at a blank text box.

Key highlights

• Next.js + Prisma + Chakra UI for the web stack
• Replicate handles DreamBooth training and Flux image generation
• Stripe integration for per-studio pricing
• S3 for image storage, Docker Compose for local dev
• OpenAI-powered prompt wizard to reduce user friction

Caveats

• Requires ~15 environment variables including AWS, Stripe, Replicate, and OpenAI keys
• The README notes you must disable TLS verification for local Docker, which is a rough edge for a security-adjacent app
• “Flux” is listed in the stack but the setup docs still reference DreamBooth training steps; the exact model flow is unclear

Verdict Worth a look if you want to bootstrap a generative AI SaaS without building payment, auth, and queue management from scratch. Skip it if you’re hoping to hack on the actual diffusion model — this stays firmly in the orchestration layer.