How I Built PhotoG: A Generative AI Tool for Realistic Product Scenes, Hitting $10k MRR in 3 Months

沈承志 — Sat, 08 Feb 2025 08:04:44 +0000

The Problem

As an e-commerce seller, I struggled to create high-quality product images that stood out. Hiring photographers was expensive, and DIY solutions looked unprofessional. I needed a way to:

Seamlessly integrate products into any scene without complex editing skills.

Generate realistic images that matched the lighting and perspective of the background.

Save time and money by automating the process.

Existing tools like Photoshop or Canva required too much manual effort, and AI tools like Midjourney couldn’t handle precise product placement.

The Build

Tech Stack:

Frontend: React + Three.js (for 3D rendering and real-time adjustments)

Backend: Node.js + Python (handles AI processing and image generation)

AI Model: Fine-tuned Stable Diffusion + CLIP (for realistic scene generation and product integration)

Database: PostgreSQL (stores user data and product libraries)

Key Features:
🖼️ Free-mode & Precise-mode: Choose between quick scene generation or detailed product placement.

🎨 Inpainting: Drag and drop products into any scene, with automatic lighting and perspective adjustments.

📦 Product Library: Upload your own products (via 360° videos or 3D models) and reuse them in multiple scenes.

🔧 Real-time Controls: Adjust product size, angle, and lighting with a simple UI.

Biggest Challenge:
Making the AI understand complex lighting and perspective matching. Solved it by training on a dataset of 100k product images with annotated lighting conditions.

Growth & Revenue

Launch Strategy:

Posted a demo video on LinkedIn and Twitter, targeting e-commerce sellers (Result: 500 signups in the first week).

Partnered with Shopify communities to offer exclusive discounts (20% off for first-time users).

Monetization:

Freemium model: Free for 10 images/month, 29/monthfor100images,29/monthfor100images,199/month for unlimited usage.

Enterprise plans for agencies and large teams (custom pricing).

Current Stats:

📈 2,000 active users

💰 $10,000 MRR (80% from subscriptions, 20% from enterprise plans)

📉 Churn: 6% (focusing on improving user onboarding and support)

What Failed

Overcomplicating the UI: Early users found the 3D controls confusing. Simplified the interface and added tooltips.

Ignoring B2B Demand: Initially focused on individual sellers, but agencies and brands were willing to pay more. Pivoted to offer enterprise plans.

Next Steps

Add a Prompt-to-Image feature for users who want to generate scenes from text descriptions.

Build a marketplace for pre-made scenes and templates.

Explore partnerships with e-commerce platforms like Shopify and WooCommerce.

DEV Community: 沈承志

How I Built PhotoG: A Generative AI Tool for Realistic Product Scenes, Hitting $10k MRR in 3 Months