TaskHub – AI-Powered Product Photography Platform

Deployed Link: https://taskhub-ai-visual.vercel.app/

(NOTE: Deployed on Vercel+Docker+Render but Render sleeps backend frequently, especially ML due to heavy loading which might cause issues and Railway is paid for deployment.)

Video: Demostration video

TaskHub is a full-stack task management platform that combines workflow management with an AI-powered Product Photography Studio.

Admins can assign product photography tasks to users, while users generate professional AI-enhanced product images and submit them for review. The platform supports role-based access control, email notifications, asynchronous image generation, task review workflows, and analytics.

---

🚀 Features

Admin Features

• Create product photography tasks
• Upload product reference images
• Assign tasks to users
• Accept submissions
• Request revisions
• View all users and tasks
• Receive submission notifications

User Features

• Receive assigned tasks
• View personal task dashboard
• Generate AI product images
• Regenerate images as needed
• Task To-do List
• Submit completed work
• Receive acceptance or revision notifications

---

📸 AI Product Photography Studio

Each task requires generation of 8 images:

Type	Count
White Background	1
Theme-Based Backgrounds	2
Creative Backgrounds	2
Model Wearing (Front)	1
Model Wearing (Side)	1
Model Wearing (Close-up)	1
Total	8

Supported Variations

White Background

• Pure white (#FFFFFF)
• E-commerce ready
• Clean product extraction

Theme-Based

Examples:

• Luxury marble surface
• Velvet premium display
• High-end showroom environment

Creative Lifestyle

Examples:

• Luxury interior scenes
• Natural outdoor settings
• Editorial-style compositions

Model Wearing

• Front view
• 45° side view
• Close-up product shot

---

🏗️ Architecture Overview

┌─────────────────────────┐
│ Frontend: Next.js in TS │
└─────────────┬───────────┘
              │
              ▼
┌─────────────────────────┐
│       Flask Backend     │
└─────────────┬───────────┘
              │
     ┌────────┴────────┐
     ▼                 ▼
┌───────────┐   ┌─────────────┐
│ Supabase  │   │ Redis Queue │
│ PostgreSQL│   │ + Celery    │
└───────────┘   └──────┬──────┘
                       ▼
             ┌─────────────────┐
             │ AI Processing   │
             │ Pipeline        │
             └─────────────────┘
                       ▼
             ┌─────────────────┐
             │ Image Storage   │
             └─────────────────┘

Workflow

1. Admin creates a task
2. Product image uploaded with reference
3. Task assigned to the user
4. User receives email notification
5. User generates AI variations
6. Images stored and tracked
7. User submits task
8. Admin receives notification
9. Admin accepts or requests revision

---

🛠 Tech Stack

Frontend

• Next.js
• TypeScript
• Tailwind CSS

Backend

• Flask
• Celery
• Redis
• Docker

Database

• Supabase PostgreSQL

Authentication

• Google OAuth 2.0
• GitHub OAuth 2.0

Email Service

• Resend

AI Components

• Background Removal
• Product Extraction
• Prompt Engineering
• Image Generation Pipeline using stability AI, Cloudflare AI

---

🗄️ Database Schema

users

Stores platform users and authentication details.

Field	Type
id	UUID
email	TEXT
full_name	TEXT
role	TEXT
provider	TEXT
created_at	TIMESTAMP
updated_at	TIMESTAMP

---

tasks

Stores assigned photography tasks.

Field	Type
id	UUID
task_title	TEXT
product_name	TEXT
product_image_url	TEXT
gender	TEXT
assigned_user_id	UUID
admin_id	UUID
status	TEXT
revision_count	INT4
created_at	TIMESTAMP
updated_date	TIMESTAMP

---

generated_images

Stores AI-generated outputs.

Field	Type
id	UUID
task_id	UUID
image_url	TEXT
prompt_used	TEXT
image_type	TEXT
created_at	TIMESTAMP

---

🔄 Task Status Flow

pending
   ↓
assigned
   ↓
in_progress
   ↓
submitted
   ↓
accepted

Revision workflow:

submitted
   ↓
revision_requested
   ↓
in_progress

---

🤖 AI Approach

Stage 1: Product Extraction

The uploaded product image is isolated from its background using automated background removal techniques.

Stage 2: Product Analysis

The system extracts:

• Product category
• Product colors
• Product characteristics

Stage 3: Prompt Generation

Dynamic prompts are created using:

• Product metadata
• Image type
• Desired angle
• Quality constraints

Stage 4: Image Generation

The AI pipeline generates:

1. White background image
2. Theme variation #1
3. Theme variation #2
4. Creative variation #1
5. Creative variation #2
6. Model front view
7. Model side view
8. Model close-up

Stage 5: Storage

Generated outputs and metadata are stored and linked to the associated task in Cloudinary.

---

🔌 API Endpoints

Authentication

POST /api/auth/signup
POST /api/auth/login
POST /api/auth/forgot-password
POST /api/auth/oauth/callback
GET  /api/auth/me
GET /api/auth/users
POST /api/auth/logout

Task

POST    /api/tasks
GET     /api/tasks
PATCH   /api/tasks/:task_id/status
GET     /api/tasks/dashboard-stats
GET     /api/tasks/my-tasks
GET     /api/tasks/:task_id
POST   /api/tasks/:id/assign
PUT    /api/tasks/:id/accept
PUT    /api/tasks/:id/request-revision
PUT    /api/tasks/:id/start
DELETE /api/tasks/:id
POST   /api/tasks/:id/submit

AI Generations

GET     /api/ai/tasks/:task_id/generations
GET     /api/ai/jobs/:id/status
DELETE  /api/ai/generations/:gen_id
POST    /api/ai/tasks/:task_id/generate
POST    /api/ai/generate

---

⚙️ Local Setup

Clone Repository

git clone https://github.com/yourusername/taskhub.git
cd taskhub

Backend Setup

cd backend

python -m venv venv

source venv/bin/activate
# Windows:
# venv\Scripts\activate

pip install -r requirements.txt
python app.py

Frontend Setup

cd taskhub-frontend

npm install

npm run dev

Start Redis

redis-server

Start Celery

celery -A workers.celery_app worker --loglevel=info

celery -A workers.celery_app.celery worker --loglevel=info -P gevent

Start Backend

gunicorn "app:create_app()"

---

🔧 Environment Variables

Create a .env file using .env.example.

# /taskhub/backend

# Flask
FLASK_ENV=

# Redis (Celery)
REDIS_URL=

SECRET_KEY=

# Supabase
SUPABASE_URL=
SUPABASE_KEY=
SUPABASE_SERVICE_ROLE_KEY=

SUPABASE_JWT_SECRET=

RESEND_API_KEY=

CLOUDINARY_CLOUD_NAME=
CLOUDINARY_API_KEY=
CLOUDINARY_API_SECRET=

# Replicate (SDXL / FLUX)
REPLICATE_API_TOKEN=

CLOUDFLARE_API_KEY=
CLOUDFLARE_ACCOUNT_ID=

STABILITY_API_KEY=

FRONTEND_URL=

# taskhub/frontend
NEXT_PUBLIC_SUPABASE_ANON_KEY=
NEXT_PUBLIC_SUPABASE_URL=

NEXT_PUBLIC_API_URL=

---

🗃 Database Setup

The project currently uses SQL schema files.

database.py
routes/

Run the schema file against PostgreSQL or Supabase before starting the application.

---

📂 Project Structure

taskhub/
├── taskhub-frontend/
│   ├── app/
│   ├── components/
│   ├── lib/
│   ├── providers/
│   ├── public/
├── backend/
│   ├── routes/
│   ├── utils/
│   ├── tasks/
│   ├── workers/
│   ├── ml/
│   ├── app.py
│   ├── config.py
│   ├── decorators.py
│   ├── database.py
│   └── Dockerfile
├── .env.example
├── README.md

---

Output Images

---

⚠ Known Limitations

• AI-generated outputs may vary depending on image quality.
• Product preservation depends on the capabilities of underlying AI providers.
• Highly reflective jewelry can be challenging.
• Model-wearing generations may introduce minor visual inconsistencies.
• Processing time depends on queue load and AI provider response times.

---

🔮 Future Enhancements

• Automated quality scoring
• Batch generation
• Fine-tuned product-preservation models
• Multi-product scene generation
• Versioned image history
• Advanced analytics

---

👩‍💻 Built By

— Khushi Goyal ✨

AI-ML Developer | MERN | Full-Stack Developer
( Sky is not the limit, Limit is in our vision. )