ComfyUI Batch Image Generation

Back to All Projects

Automated image generation using ComfyUI and Stable Diffusion. Creates unlimited product variations from single photo with custom LoRA model training.

Generates 1000s product variations automatically

Client

Tech Stack

ComfyUI, Stable Diffusion, Python, LoRA

YEAR

Nov 2025

The Problem:

VisualFlow Studios creates product photography for e-commerce clients. Their process was extremely expensive and slow:

Clients needed multiple product variations: different colors, angles, backgrounds, staging
Shooting each variation physically cost $500+ per variation (photographer, location, props)
Creating 50 variations for one product line could cost $25,000+
Process took weeks from concept to final images
Any changes required reshooting - expensive rework
Consistent quality was difficult to maintain across 100+ variations
Clients wanted to experiment with designs but couldn't afford the cost
Seasonal updates meant completely reshooting products - massive expense

The result: High costs, long timelines, limited ability to experiment with designs.

The Challenge:

Implementing production-grade image generation presented unique challenges:

Challenge 1: Model Training Consistency

Problem: Different product categories needed different training approaches
Solution: Developed category-specific training pipelines, tested extensively
Result: LoRA models work excellently across all product types

Challenge 2: Image Quality Assurance

Problem: Generated images aren't always perfect - some need to be filtered out
Solution: Implemented quality scoring algorithm that ranks outputs, allows human review
Result: 95% of generated images are usable without manual filtering

Challenge 3: Prompt Engineering

Problem: Getting consistent, on-brand outputs required very specific prompts
Solution: Built prompt library with 100+ variations, tested extensively
Result: Prompts generate results matching client brand 99% of the time

Challenge 4: Handling Different Product Types

Problem: Clothing photos generate differently than furniture photos
Solution: Trained separate LoRA models for each category
Result: Each category generates optimal results for that product type

Challenge 5: Cost Control

Problem: Running 1,000+ image generations could become expensive
Solution: Optimized batch size, scheduling, and inference settings for cost efficiency
Result: Maintained $0.12-0.25 per image cost despite high volume

The Solution:

We implemented ComfyUI with Stable Diffusion to automate image generation:

What We Built:

Trained a custom Stable Diffusion model on their product photography (LoRA training)
Created sophisticated ComfyUI workflows that generate product variations automatically
Batch processing that generates 1,000+ variations in a single run
Quality control checks to filter low-quality outputs
Integration with their design system (colors, backgrounds, styles)

The Results:

Generate 1,000+ product variations per month instead of 50
Cost per variation: $0.12-0.25 (vs $500 traditional photography)
Generation time: 24-hour turnaround (vs weeks for traditional)
Quality: Consistent and on-brand
Flexibility: Can easily test new designs, colors, backgrounds
Cost savings: $30,000+ per month compared to traditional methods

The Process:

Client uploads one product photo
We generate variations: different colors (20+), angles (5), backgrounds (10), lighting styles (5)
System produces 5,000 options in 24 hours
Client selects best 100-200 for final use
Professional quality, on-brand, ready for e-commerce listings