Self Hosting Comfy UI

Experimenting w Image Generation

Interfaces like A1111, Comfy UI have spearheaded the image generation space, with creators being given a powerful tools to unleash their creativity.

Models like the stable diffusion model, Flux make image generation the new cool thing kids like to do. Social Media is already flooded with AI generated content. Apart from main models, LoRa models provide ways to manipulate images in hundreds of ways.

Although Comfy UI Cloud exists and allows you to use their service, there are usually waitlists and limits/ credits that hinder the experience, especially if you’re trying out different styles. This is where the community version of LoRa comes in. You can choose to either set this up locally on your machine, if your hardware supports it - 32 OR 64GBRAM. Or you can deploy this on a cloud server.

Nov 2025 Update

I recently found out about a Docker way to setup Comfy UI. ALthough I haven’t done this myself yet, I think its the best way to setup rather than setting up everything manually like I’ve done below.

Here’s the link - Run ComfyUI in Docker Quick and Easy

TODO (This section is still being )

• setup using official guide

• troubleshooting

• gotchas

• screenshots

SCREENSHOTS

Good Practices To Save Costs

I learnt this the hard way. AWS bills are a killer, even for passion projects, especially around AI.

• Setup alarms and budgets before you start.

• Setup idle time shutdowns, say 15mins.

• Snapshot regularly and delete resources (terminate instance, detach EBS, remove IPs) as they incur heavy charges.

• Play around with different instance sizes to see which works for your use case.