LOCAL AI

Local AI Product Image Generation

A desktop app that locally generates product images blocked by cloud content filters.

Role
Developer
Status
In Development
Platform
Windows · macOS
Industry
Textiles · Lingerie
Local AI Product Image Generation — interface and work in progress
OVERVIEW

How Local AI Image Generation works

Local AI Product Image Generation is a desktop application that turns product photos into catalog and campaign visuals entirely on the company’s own computer. It is designed specifically for lingerie, swimwear, and textile products that often get blocked by the content filters of cloud-based generative AI tools. The project is currently in development.

The problem: content filters and data leaving the company

Due to automated content policies, cloud-based image generation tools frequently reject completely legitimate product photos such as lingerie and swimwear. On top of that, photos of unreleased products and model data end up uploaded to third-party servers. Together, these two issues make it difficult for brands in this sector to benefit from generative AI.

The solution: an end-to-end local generation workflow

  • Product input: the product photo and key attributes are added to the app.
  • Template: a brand-specific style, scene, and lighting template is selected.
  • Generation: a ComfyUI workflow generates the visual using local diffusion models.
  • Output: color and background variations are generated in batches and exported.

All of these steps can run without an internet connection; product data never leaves the company’s computer.

Technical architecture

The desktop layer is built with Tauri and Rust, keeping the application lightweight and cross-platform on Windows and macOS. The interface is built with Vue 3 and Pinia, designed to be simple enough for non-technical users. Image generation runs on ComfyUI using SDXL and brand-specific fine-tuned (LoRA) models, while project and generation records are stored locally in an SQLite database.

Queue management and overnight batch generation

Generating variations for hundreds of products can take hours. The application places jobs into a queue and can run batch generation when the computer is idle—for instance, overnight. The ComfyUI queue script I wrote for this task is shared on GitHub.

Further reading

I shared the background and experiments behind this approach in my note on product image generation with local AI. If you are considering a similar setup for your own products, take a look at my desktop applications and local AI service.

01 — WHY?

Where the project started

When lingerie, swimwear, and textile brands try to create product visuals with cloud-based generative AI tools, they run into content policies—and their product and model data leaves the company. This application brings the entire generation process onto the company’s own computer.

Avoid reliance on cloud content filters
Keep product data in-house
Easy to use for non-technical teams
02 — HOW DOES IT WORK?

Architecture and flow

01
Product input
Product photo and attributes
02
Template
Brand-specific style and scene
03
ComfyUI
Local diffusion workflow
04
Output
Batch variations and export
Desktop
TauriRust
Interface
Vue 3Pinia
Models
ComfyUISDXLLoRA
Data
Local diskSQLite
03 — HIGHLIGHTS

What does it offer?

  • Generation workflow that works offline
  • Brand-specific style, scene, and lighting templates
  • Batch color and background variations from a single product
  • Overnight batch generation with queue management
  • Simple desktop interface requiring no technical expertise
HAVE A SIMILAR NEED?
Want to generate product visuals without your data leaving your company?
Related service
OTHER PROJECTS