How To Generate AI Images Locally on a Gaming PC with a Nvidia GPU

I’ve been experimenting with AI image generation since Dall-E 2. Mostly this has been weird art creation, but specific to this challenge I’ve got more and more interested in business practical uses for AI image generation. 

As part of my AI Directory Maker I needed to get AI to design several image assets.

Right now the API options for AI image gen are getting good at this (see my post: Which AI Models Make the Best Logo’s in 2025?)

But really good AI image generation costs $. Also, I happened to have a gaming rig under my desk with a meaty, (though old), Nvidia 1080 Ti card, which just about passes the bar to run the smaller image generation models. Let’s see if we can trade solar power off my roof for crisp AI generated images.

Required PC Hardware for AI Image Generation

There are lots of PC setups that will run AI image generation models. The consensus seems to be that Nvidia GPUs are kind of essential, and obviously you’ll have a much easier time if you’re running a decent recent motherboard, CPU, and have lots of fast RAM.

Luckily for gamers; if you can run games at high FPS and resolution, the chances are you can repurpose your box for AI image gen. (I had to quit gaming because of migraines, so this reuse is perfect!)

Here’s what I ran all these tests on:

• AMD Ryzen 9 5950X
• 32 GB DDR4
• Nvidia GTX 1080 Ti
• Asrock X570M Pro4 board

Aside: I picked up my 1080 Ti years ago before crypto and AI had smashed the market, it’s still worth more than what I bought it for 🫠 – if you’re doing this seriously you’d be wanting H100’s, (but you can also rent them fairly cheaply).

Tooling your PC to Generate AI Images

I read a lot about this before I started and almost unanimously people suggest ComfyUI.

Installing ComfyUI was easy. After it downloaded the models I wanted to use (Flux Dev and Flux Schnell) which were 16GB each, it worked via the UI app first time.

Here’s my first image gen workflow:

It produced a passable image:

I messed around with Flux AI image gen models a fair bit and even though they’re less amazing than OpenAI’s 4o and Google’s Gemini Flash 2.0, they still are very capable.

Depending on your use case it may well be viable to use your gaming PC to generate images.

My Use Case for Gaming PC Generated AI Images

I’ve been exploring another directory alongside my AI Directory Maker which I’ll announce soon. It’s kind of a monster, with a lot of pSEO generated pages to begin with.

Programmatic SEO is still a great way to boost your websites search surface and I have regularly used it to produce pages en masse for websites.

(The caveat is that it’s very important to make unique, highly useful, genuine content – not spam slop. The latter will very quickly get your site slapped by Google.)

This new directory site lists an awful lot of services and objects relating to AI and business.

It’s a database I’ve been sculpting for a while which totals 20k+ lines for this first launch, (I’ll be slowly drip-feeding content out as I work my way through to ensure quality and to not swamp the search engines.)

I need unique images to add value to each page.

Generating 40k images via OpenAI or Flash 2.0 would not be cheap, (like $4k+ not cheap). That’s using OpenAI’s batching process too, so it’s not even generating images when I ask for them.

So… down the solar-powered-gaming-pc-running-ai-image-gen rabbit hole we go.

Creating AI Image Generation Workflows in ComfyUI

ComfyUI is a drag and drop AI model running workflow tool. Not to be confused with all of the other AI workflow tools I’ve been talking about here; ComfyUI takes steps to run prompts on AI models directly.

Therefore if we want ComfyUI to push our prompt through a model like Flux Dev we need to give it a workflow.

Here’s a simple example:

Now all we really care about is the prompt, but we can play with different models, change the various weights, and add a negative prompt if we want to finesse.

Either way once we have our flow working in the ComfyUI app we then need to export the workflow into a JSON file so we can pump that out for each image we need via script.

To do this we go to ‘Workflow’ -> ‘Export API’ from the top menu.

This gives us a JSON file we can then import, modify, and send via CURL.

Batching Hell

ComfyUI has a nice API server you can run so that you can theoretically just call it with workflow JSON to generate images.

Given how easy it was to set up ComfyUI to work via UI I expected to be able to call it via some python or command line and then just wire up my DB to a script which stacks up these image generation runs and logs the output.

I was wrong.

First I decided I wanted to write a script in php, because later it’d be easier to integrate it into my business API. This failed:

1. When I used php to parse the JSON, switch out the prompt and filename prefix, the reformatted JSON was not in the same format
2. …so I tried to force various format changes
3. …then I tried to inject the prompt and filename strings literally
4. …then Cursor.ai got lost down many rabbit holes fixing “bugs” in ComfyUI’s python

Needless to say it was frustrating all round.

So I switched back to python.

Using python to automate ComfyUI is far easier. 

In theory it shouldn’t matter – given that ComfyUI is being called via CURL request. Maybe it was just my tired brain that first day.

Anyhow via python ComfyUI took the exported, tweaked, workflow template JSON easily and just started making images. After that I tweaked the script to:

• Take a CSV file (db export) and run lines x to y
• Generate a seo-friendly filename prefix
• For each line reload the workflow JSON, and modify the prompt/filename prefix
• Queue up ComfyUI to generate the AI image

I was overall happy with the results, given that just a few years ago none of this was really possible. I much prefer OpenAI’s 4o – but given that 1) it’s expensive, 2) it’s not even available via API yet, I can live with these simpler images.

Note: You can also use the ComfyUI-to-Python-Extension to generate executable python scripts directly from ComfyUI workflows. I personally didn’t find that this worked as well as simply queueing up via API.

Final System for Generating AI Images with Gaming PC

Here’s my final tooling I am using to generate the 40k images. I’ve written this as a kind of guide for anyone wanting to generate images with their gaming PC, but I’m not sure what use case would be practical here apart from my obscure pSEO things.

1. Verify Gaming PC is High Enough Spec

• Ideally you have a fast Nvidia GPU with lots of RAM
• Ideally you have a fast CPU, 32GB+ of fast RAM

2. Install ComfyUI for Image Gen

I ran this on Windows 11, ComfyUI will work on other OS’s but here I’ll refer to Windows tooling.

1. [Optional] Install ComfyUI as an app from comfy.org (the UI installed app is easier to build your workflow in before exporting)
2. git clone the ComfyUI repo into a directory like C:/AI 

git clone https://github.com/comfyanonymous/ComfyUI.git C:\AI\ComfyUI

3. Downgrade your python if it’s > 3.11.9 (I had a lot of issues running newer python)
4. [If Nvidia GPU] Install latest version of CUDA
5. Install PyTorch

pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu126

6. [If you’re me] Uninstall PyTorch and reinstall a bunch of times because you didn’t read the guide properly
7. Run pip install

pip install -r requirements.txt

8. Start venv

python -m venv venv
venv\Scripts\activate

9. Run the ComfyUI server

python main.py

… if that all worked you should see a nice server message like this:

… and you should be able to see a similar UI to the full ComfyUI app at http://127.0.0.1:8188

3. Write & Run Queueing Python Script to Populate ComfyUI queue

That means you’re good to go. You can see the script, workflow template, and CSV file I wrote for this here in the ProfitSwarm git repo – but you might want to tweak it for your use case.

• For this script to work you’ll have to have a running ComfyUI server which you can see live when you go to http://127.0.0.1:8188
• You’ll need to replace the CSV with data relevant to your use case
• You might want to tweak or replace the workflow_template.json (e.g. change output resolution or AI model)

Let me know if you use it, or have any issues generating big batches of AI images from it. 

4. Notes on Generating AI Images This Way

• Because ComfyUI is loading the AI model to GPU memory it will always take a while to generate the first image, but so long as the server stays up all further images will be much faster. 
• I found that if you switch models (e.g. from Flux Dev to Flux Schnell) mid queue it tends to crash the server. So always use one model per queue else you’ll check back in to see a dead server and AI images left ungenerated.

5. Different AI Image Generation Models

There are many AI models you can download and use in ComfyUI but I found that it can be a deep hole to fall into.

To get started quickly I recommend trying Flux models:

• Flux Dev (flux1-dev-fp8.safetensors) for higher quality output
• Flux Schnell (flux1-schnell-fp8.safetensors) for faster output

If you come across any excellent models please do share them in the comments below.

What I Learned Trying AI Image Gen

It’s awesome what AI can do from your browser or via API, but it’s a whole different vibe when it’s all running locally on your PC. The internet could die tomorrow and you’d still have this running without issue.

Now of course it’s incredibly unlikely that’ll happen, but if I’d used APIs for this I’d have run into the thousands of dollars for AI image generation.

In time I suspect these models will get better and better, cheaper and cheaper, and we won’t have much practical need for approaches like this, but it was fun to get some business use out of this old gaming rig.

Here are my key takeaways:

• Prompts matter – Be descriptive, but avoid over explaining. Too long prompts can muddy/confuse the model more than they help. Use negative prompts to your advantage
• AI models can run fine on your gaming PC but speeds are much less than cloud provided AI
• It’s kind of cool generating AI images for pSEO on a gaming PC running off solar power – with no net connection needed!

Other Approaches to Generating AI Images En Masse

If you don’t have a gaming PC with a strong GPU you can still get this work done, it will cost you a lot more, (but be faster and better quality). Here are the avenues I’d explore without the gaming PC:

• Use Google Gemini 2.0 Flash via API (batching saves 50-70%)
• Use OpenAI 4o (via macro) or DALL-E 3 (use HD) (batching saves 50%)
• Rent a H100 and run your AI image gen yourself (probably cheapest but more technical)

How are You Using AI to Generate Images?

What images are you finding yourself making with AI? Business stuff like logo’s, ads, or sharing images? Please comment below and share what you’re using AI to generate.

…and if you know someone who’s into AI image generating, send them this post!