Fooocus AI is the Stable Diffusion XL image generator that hides most of the knobs: write a prompt, pick a style, click Generate. It needs an Nvidia GPU, which is exactly what most laptops lack. Meetrix's AMI runs it on AWS instead, with Fooocus and Fooocus-API installed on Ubuntu 22.04 and a g4dn.2xlarge GPU instance recommended.
Know one thing before you invest time in it: Fooocus is in limited long-term support. Its maintainer fixes bugs but has no plans to adopt newer model architectures, so it stays an SDXL tool. For SDXL work that's fine, and it's still one of the easiest ways to get good images quickly. If you need the newest models, read our Fooocus vs Automatic1111 vs ComfyUI comparison first.
Fooocus AI on AWS: guides, video and listing
Quick Video Guide
Deploy the Fooocus AI image generator on AWS
- Check your GPU quota first. In Service Quotas, look at "Running On-Demand G and VT instances" for your region; a g4dn.2xlarge needs 8 vCPUs of it. Many accounts start lower. Our vCPU quota guide walks through the request.
- Open the Fooocus listing on AWS Marketplace, subscribe and accept the terms.
- Choose "CloudFormation Template" and your region, then launch the stack.
- Fill in
Admin Email,FooocusDomainName(a Route 53 domain gets SSL automatically) andFooocusInstanceType(g4dn.2xlarge recommended). Submit and wait 5-10 minutes. - Open
DashboardUrlIpfrom the Outputs tab. A 502 in the first few minutes means Fooocus is still loading; give it about 5 minutes. - Type a prompt and click Generate. The first image after a start is the slowest, while the models load into GPU memory; the next ones come much faster.
Full screenshots, manual SSL and the API endpoints are in the Fooocus developer guide. Running on Google Cloud instead? See Fooocus on GCP.
Fooocus minimum VRAM and the right instance
Fooocus's README asks for at least 4 GB of VRAM on Nvidia RTX 20, 30 or 40 series cards, 8 GB on GTX 10 series, and 8 GB of system RAM with swap turned on. Those are minimums for a desktop. On AWS the g4dn family is the usual fit:
| Instance | GPU | vCPU / RAM | Use it for |
|---|---|---|---|
| g4dn.xlarge | Nvidia T4, 16 GB | 4 / 16 GiB | One person, occasional images |
| g4dn.2xlarge (recommended) | Nvidia T4, 16 GB | 8 / 32 GiB | A small team, or the API under light load |
Going bigger within g4dn adds CPU and RAM, not GPU speed: every size from xlarge to 8xlarge has the same single T4. If an app needs more images per minute, run a second instance rather than a larger one.
What Fooocus AI costs on AWS
The Meetrix software fee is $0.019 per hour on a g4dn.2xlarge. The GPU instance itself costs far more than that, so check the EC2 on-demand price for your region before you launch. The habit that saves the most money is simple: stop the instance when nobody's generating. A stopped instance costs only its EBS storage, and starting it again takes a couple of minutes. Pair that with an AWS billing alarm so a forgotten GPU doesn't run all weekend.
Fooocus compared with other image generators
Fooocus is the simplest of the self-hosted SDXL tools: few settings, good defaults, styles and inpainting built in. Automatic1111 gives you far more options and extensions. ComfyUI builds images from node graphs and gets new models first. We compare all three in detail in Fooocus vs Automatic1111 vs ComfyUI, and the wider field in our roundup of self-hosted AI image generators.
Frequently Asked Questions
What is Fooocus AI?
Fooocus is a free, open-source (GPL-3.0) image generator built on Stable Diffusion XL. You type a prompt and it handles most settings itself, closer to Midjourney's simplicity than to node-based tools like ComfyUI, while still running on your own GPU.
Is it Fooocus or Foocus?
Fooocus, with three o's. People often search for Foocus or Focus AI image generator, but the project on GitHub, by lllyasviel, is spelled Fooocus.
What is the Fooocus minimum VRAM?
Fooocus's README lists 4 GB of VRAM on Nvidia RTX 20, 30 and 40 series cards, 8 GB on GTX 10 series cards, and at least 8 GB of system RAM with swap enabled. The recommended g4dn.2xlarge has a 16 GB Nvidia T4, comfortably above that.
How much does Fooocus on AWS cost?
Two parts: the Meetrix software fee, $0.019 per hour on the recommended g4dn.2xlarge, and the EC2 charge for the GPU instance itself, which is most of the bill. Stop the instance when nobody's generating and you stop paying for compute.
Why can't I launch a g4dn instance?
Many AWS accounts start with little or no quota for G-family GPU instances. Request an increase for Running On-Demand G and VT instances in Service Quotas before you launch; approval can take a day.
Can I run Fooocus in Google Colab instead?
Fooocus's README links a Colab notebook, and it's fine for trying the tool. But sessions time out, free GPUs aren't guaranteed, and nothing stays running for a team. A GPU instance you control is the better fit for regular or API use.
Does the AMI include an API?
Yes. It ships with Fooocus-API alongside the web interface, so other applications can generate images over HTTP. The developer guide lists the endpoints.
Run Fooocus AI on Your Own AWS GPU
Launch Fooocus and Fooocus-API on a g4dn GPU instance in your own AWS account, and stop it whenever you're done generating.
Deploy Fooocus on AWS Marketplace