PowerPod GPUsUse casesCase studiesGuidesBlogFor collegesHelp Sign inCreate account

Example project · for online stores and marketing teams · built from today’s prices and our typical run times; not a customer story. Prices updated 21 November 2026.

AI product photography: 1,000 product images for USD 0.72

Example project: generate a thousand product and lifestyle images with Stable Diffusion XL in ComfyUI on a rented GPU, at today's prices.

Key takeaways
USD 0.72Cost today
USD 0.01Per image
2.4 hGPU time
ComfyUI, Stable Diffusion XLSoftware
Start this projectPrice my own job

The challenge

An online store needs lifestyle pictures for hundreds of products and fresh banners every season. Photo shoots are slow and expensive, and image tools on a subscription limit how many pictures you can make.

Why a rented GPU

How it works, step by step

  1. Rent a GPU with ComfyUI and open it in your browser.
  2. Optionally train a small LoRA on 15 to 30 photos of a product, so it looks the same in every picture.
  3. Queue prompts for each product and scene, and download the results.
  4. Check every image before you publish it, then destroy the machine.

Which GPU to choose

The cheapest way today is RTX 3090: 2.4 hours for USD 0.72.

The quickest, H100 SXM, takes 0.8 hours for USD 2.99.

FLUX makes sharper text and hands but takes about three times as long per image, and wants 24 GB or more.

What it costs today

StepGPUTimeCost
Generate 1,000 images (Stable Diffusion XL, 1024×1024)RTX 30902.4 hUSD 0.72
The same, on the quickest cardH100 SXM0.8 hUSD 2.99

GPU time at today’s lowest price on a verified host, from typical run times: your own job may run faster or slower, and prices move with the market. Disk storage adds a little and is shown before you rent.

What you end up with

Common mistakes to avoid

Keep the bill where you want it

Start this projectPrice my own job

Questions

How much does AI product photography cost?

On PowerPod today about USD 0.01 an image: 1,000 images for about USD 0.72 of GPU time.

Can the products look like the real thing?

Train a LoRA on photos of the product first (Kohya, about an hour on one GPU), then use it in every prompt. Check the results before you publish them.

Can we use the images commercially?

Stable Diffusion XL's licence (CreativeML Open RAIL++-M) allows commercial use of what you generate, within its use restrictions. Check the licence of any other model you use.

Read next

More case studies

Train a chatbot on your own dataFor SaaS and support teamsTranscribe hours of audio with WhisperFor call centres, podcasts and mediaA self-hosted LLM API for a client's appFor AI agencies and app developersCloud GPUs for students: an AI course labFor colleges, bootcamps and trainers

These are example projects built from today’s prices and the typical run times on our GPU pages, not customer stories. When customers agree to share theirs, they will appear here. Written by the PowerPod team.