AWS EC2 Deployment¶

Deploy Image Editor to AWS EC2 using Terraform.

Overview¶

The Terraform configuration provisions:

VPC with public subnet
Security groups for web traffic
EC2 instance with nginx
Automated application deployment

Prerequisites¶

AWS account with appropriate permissions
Terraform 1.0.0+
AWS CLI configured with credentials

Quick Start¶

1. Configure Variables¶

cd terraform
cp terraform.tfvars.example terraform.tfvars

Edit terraform.tfvars with your settings:

aws_region    = "us-east-1"
instance_type = "t3.micro"
environment   = "dev"

2. Initialize Terraform¶

terraform init

3. Review Plan¶

terraform plan

4. Apply¶

terraform apply

Confirm with yes when prompted.

5. Access Application¶

Get the application URL:

terraform output app_url

6. Connect to Instance (Optional)¶

Connect via AWS Systems Manager Session Manager:

terraform output ssm_session_command

Or use the AWS Console: EC2 → Select instance → Connect → Session Manager tab.

Configuration Options¶

Variable	Description	Default
`aws_region`	AWS region	`us-east-1`
`instance_type`	EC2 instance type (non-GPU)	`t3.micro`
`environment`	Environment name	`dev`
`enable_qwen_image_edit`	Enable Qwen Image Edit diffusion model	`false`
`qwen_model_variant`	Model variant: `full` or `fp8`	`full`
`gpu_instance_type`	GPU instance type	`g5.2xlarge`
`diffusion_api_port`	Diffusion API port	`8000`

Architecture¶

┌─────────────────────────────────────────────────────────┐
│                         VPC                             │
│  ┌───────────────────────────────────────────────────┐  │
│  │                  Public Subnet                     │  │
│  │  ┌─────────────────────────────────────────────┐  │  │
│  │  │         EC2 Instance (GPU optional)         │  │  │
│  │  │  ┌─────────┐    ┌───────────────────────┐  │  │  │
│  │  │  │  nginx  │────│     Image Editor      │  │  │  │
│  │  │  │ (:80)   │    │    (React Frontend)   │  │  │  │
│  │  │  └─────────┘    └───────────────────────┘  │  │  │
│  │  │                                             │  │  │
│  │  │  ┌─────────────────────────────────────┐   │  │  │
│  │  │  │   FastAPI Diffusion Server (:8000)  │   │  │  │
│  │  │  │   (Qwen-Image-Edit-2511 model)      │   │  │  │
│  │  │  └─────────────────────────────────────┘   │  │  │
│  │  └─────────────────────────────────────────────┘  │  │
│  └───────────────────────────────────────────────────┘  │
│                          │                               │
│                 Internet Gateway                         │
└──────────────────────────┼───────────────────────────────┘
                           │
                      Internet

Qwen Image Edit (AI Feature)¶

Enable AI-powered image editing with the Qwen-Image-Edit-2511 diffusion model.

Enable AI Image Editing¶

enable_qwen_image_edit = true
gpu_instance_type      = "g5.2xlarge"
qwen_model_variant     = "full"

GPU Instance Options¶

Instance	GPU	VRAM	vCPU	RAM	Cost/hr
g5.xlarge	A10G	24GB	4	16GB	~$1.01
g5.2xlarge	A10G	24GB	8	32GB	~$1.21
g5.4xlarge	A10G	24GB	16	64GB	~$1.62

API Usage¶

After deployment, the Diffusion API is available:

# Get API URL
terraform output diffusion_api_url

# Health check
curl http://<ip>:8000/health

# Edit an image
curl -X POST "http://<ip>:8000/edit" \
  -F "image=@input.jpg" \
  -F "prompt=Add a sunset background" \
  -o output.png

Model Variants¶

full: Best quality, ~40GB download, requires 24GB VRAM
fp8: Faster inference, ~20GB download, reduced VRAM usage

For detailed usage instructions, prompt examples, and best practices, see the AI Image Editing Guide.

Security¶

Security Groups¶

HTTP (80) - Open to all
HTTPS (443) - Open to all

Instance Access¶

Instance access is provided via AWS Systems Manager Session Manager, which offers:

No inbound ports required (no SSH port 22 exposed)
IAM-based authentication and authorization
Session logging and auditing capabilities
No need to manage SSH keys

Best Practices¶

Use IAM policies to control Session Manager access
Enable AWS WAF for additional protection
Configure SSL certificates via ACM

Maintenance¶

Updating the Application¶

Build new version locally
Upload to S3 or rebuild on instance
Restart nginx if needed

Scaling¶

For higher traffic:

Use a larger instance type
Add Application Load Balancer
Configure Auto Scaling Group

Cleanup¶

Remove all resources:

terraform destroy

Data Loss

This will delete all AWS resources. Ensure you have backups if needed.

Troubleshooting¶

Cannot Connect via Session Manager¶

Ensure the SSM agent is running on the instance
Verify IAM permissions include ssm:StartSession
Check that the instance has internet access (required for SSM)
Install the Session Manager plugin for AWS CLI: Installation Guide

Application Not Loading¶

Check nginx status: sudo systemctl status nginx
View nginx logs: sudo tail -f /var/log/nginx/error.log
Verify build files exist in /var/www/html

Diffusion Server Issues¶

Check service status: sudo systemctl status diffusion-server
View logs: SYSTEMD_LESS=FRXMK sudo journalctl -u diffusion-server -f
Check GPU availability: nvidia-smi
Verify model loaded: curl http://localhost:8000/info
Model download can take 10-30 minutes on first startup

More Information¶

See the Terraform README for detailed configuration options.