Own AI Product
Children's drawings become custom 3D figurines for painting - powered by a multi-model AI pipeline with Google Gemini and Hunyuan3D, orchestrated by FastAPI and Redis.
Category
D2C E-Commerce
Technology
Multi-Model AI, 3D Printing
Status
Live
Website
doodlhero.com
Before
After
MVP Development Time
8 Weeks
Core Team
3 People
Time to First Sale
4 Weeks
Multi-Model Pipeline
3 AI Models
Stripe, Cloudinary, B2 & more
6 Integrations
Evolution: V1 → V2
Version 1 — Semi-Automated
Orders were stored in Monday.com, 3D models manually created and exported in GLTF format. Turnaround took several days per order.
Version 2 — Fully Automated
External AI APIs automatically generate 3D models in OBJ/MTL format. Job polling enables real-time feedback with instant 3D preview and change requests.
MVP & Validation
Manual pipeline, Monday.com, first sales. Product-market fit confirmed.
AI Integration
Google Gemini & Hunyuan3D connected. First automated 3D generation.
Real-Time Pipeline
Redis job queue, Three.js preview, before/after comparison live in browser.
Scaling
Stripe Payments, Cloudinary CDN, Backblaze B2 Storage. Fully automated & profitable.
The Problem
Children's drawings pile up - too precious to throw away, but too many to keep. A physical solution that preserves memories was missing.
The magical time when children draw uninhibitedly never comes back. Parents want to capture this phase - not just digitally, but tangibly.
Families seek shared activities beyond tablets and smartphones. Creative quality time instead of passive consumption.
Standard plastic toys end up in the corner after a week. Parents and grandparents are looking for meaningful gifts.
The Product
Step 1
Step 2
Step 3
How It Works
Next.js + Cloudinary
Parents upload a drawing via the Next.js frontend. The image is optimized through Cloudinary and sent to the FastAPI backend, which creates an asynchronous job in the Redis queue.
Gemini -> PrunaAI -> Hunyuan3D
Google Gemini optimizes the image, PrunaAI segments the contours, Hunyuan3D-2/Pro generates the 3D model as OBJ/MTL. The client polls job status in real-time via ARQ.
Three.js / React Three Fiber
The generated 3D model is rendered live in the browser via React Three Fiber - with rotation, zoom, and before/after comparison. Customers can request changes before purchasing.
Stripe + Make.com
Stripe handles payment, Make.com webhooks trigger the 3D printing workflow. The figurine is printed from Bio-PLA and shipped as a complete set with paints.
What Makes DoodlHero Special
Three specialized AI models (Google Gemini, PrunaAI, Hunyuan3D) work together for optimal 3D results.
Free 3D preview before purchase. Payment only when the customer is satisfied - zero risk.
Not just a gift, but shared quality time. Paint together, be proud together.
3D figurine (12-15 cm), 6 acrylic paints, 2 brushes. All inclusive, no hidden costs.
Bio-PLA from cornstarch, non-toxic paints, minimal packaging. Handmade in Germany.
FastAPI + Redis/ARQ job queue on Docker/Railway. Parallel jobs, real-time polling, zero-downtime deployments.
Traction
120+
Figurines Sold
4.9/5
Customer Rating
~2 min
Real-Time AI Pipeline
V1 -> V2
Manual to Fully Automated
"The painting was the highlight! My daughter was so proud when she finished HER figurine."
- Katharina M., Mother of Emma (5)
"We gave DoodlHero as a birthday gift. My grandson hasn't let go of his figurine since."
- Thomas R., Grandfather
"Three orders in two months. Once you start, you want every drawing as a figurine."
- Sarah L., Repeat Customer
"The painting was the highlight! My daughter was so proud when she finished HER figurine."
- Katharina M., Mother of Emma (5)
"We gave DoodlHero as a birthday gift. My grandson hasn't let go of his figurine since."
- Thomas R., Grandfather
"Three orders in two months. Once you start, you want every drawing as a figurine."
- Sarah L., Repeat Customer
Architecture & Technology
Three specialized AI models work together: image optimization, image-to-3D conversion, and mesh refinement. Orchestrated by a FastAPI backend architecture with asynchronous job processing.
Next.js Frontend
React-based upload interface with Framer Motion animations. Image is sent to FastAPI.
Google Gemini
gemini-2.5-flash-image optimizes and prepares the child's drawing for 3D conversion.
PrunaAI
Image processing via Replicate API. Segmentation and contour optimization of the drawing.
Hunyuan3D
Tencent Hunyuan3D-2 & Pro via Replicate. Image-to-3D conversion, output as OBJ/MTL.
Three.js Viewer
React Three Fiber renders the 3D model live in the browser. Before/after comparison and change requests.
Three specialized models in series: Google Gemini (image optimization), PrunaAI via Replicate (segmentation), Hunyuan3D-2/Pro (3D synthesis). Each model fulfills one task - the result is better than any single model could achieve.
3 AI Models
Gemini -> PrunaAI -> Hunyuan3D
FastAPI (Python 3.12) with Redis and ARQ for asynchronous job processing. The client polls job status in real-time - no waiting without feedback. Deployed on Railway in Docker containers.
Real-Time Polling
Redis + ARQ Job Queue
Next.js with Tailwind CSS and Framer Motion for fluid UX. Three.js via React Three Fiber renders the generated 3D model live in the browser - with rotation, zoom, and before/after comparison.
Live 3D Preview
Three.js / React Three Fiber
Lessons Learned
No single AI model could reliably convert children's drawings to 3D. Only the combination of Google Gemini (image optimization), PrunaAI (segmentation), and Hunyuan3D (3D synthesis) produced consistently good results. Orchestrating specialized models is often more effective than a single general-purpose model.
V1 was deliberately semi-automated: manual 3D modeling via Monday.com, multi-day turnaround. It worked and proved the market exists. Only with that certainty did we build the fully automated pipeline. MVP first, scaling second.
Switching from "we'll get back to you in 3 days" to a live 3D preview via Three.js in the browser was a game-changer. Customers see their result instantly, can request changes, and buy with confidence.
Instead of building everything: Stripe for payments, Cloudinary for CDN, Backblaze B2 for storage, Make.com for webhooks, Monday.com as CRM. Focus on core competency (AI pipeline) while specialized services handle the rest better.
What This Means for Your Business
At DoodlHero
becomesfor you
becomes for you
For Your Business
Invoices, delivery notes, contracts: automatically captured, classified, and transferred to your system.
At DoodlHero
becomesfor you
becomes for you
For Your Business
Order, production, shipping - without manual steps. The same principle for your order chain.
At DoodlHero
becomesfor you
becomes for you
For Your Business
Proof of concept in weeks, not months. No budget risk until value is proven.
Frequently Asked
DoodlHero uses a three-stage multi-model pipeline: Google Gemini (gemini-2.5-flash-image) for image optimization, PrunaAI via Replicate for segmentation and contour optimization, and Tencent Hunyuan3D-2 and Hunyuan3D Pro for image-to-3D conversion in OBJ/MTL format.
15 minutes. Free. No strings attached.
Not a sales pitch. Promised.
More Projects
AI Chatbot
An AI assistant for customer support - trained on company data, deployed in under 2 weeks.
Generative AI
AI-generated tattoo designs based on user prompts. 10,000+ designs in the first month.
Marketplace
Tattoo marketplace for discovering and booking artists. Google Maps integration, real-time search.