MyFit

I wanted a safe and private way to see how clothes will fit with my existing wardrobe before buying them online, so I quickly put this together to run on my 3090.

Virtual try-on functionality powered by IDM-VTON.

Features

• Async job queue for GPU inference
• Prometheus metrics and OpenTelemetry tracing
• Kubernetes-ready health probes
• Request correlation IDs
• Structured JSON logging
• NixOS module and Docker deployment

Requirements

• NVIDIA GPU with CUDA support
• ~12GB VRAM (float16) or ~24GB VRAM (float32)
• Python 3.11+ or Nix/Docker

Quick Start

Docker

# Standard (requires nvidia-container-runtime)
docker run --gpus all -p 8000:8000 ghcr.io/clairesrc/myfit:latest

# NixOS (uses CDI)
docker run --device nvidia.com/gpu=all -p 8000:8000 ghcr.io/clairesrc/myfit:latest

Nix

nix develop  # Enter dev shell
python -m myfit.api.main --host 0.0.0.0

pip

pip install -e ".[all]"
python -m myfit.api.main

The API is available at http://localhost:8000 with docs at /docs.

API

Submit Try-On

curl -X POST http://localhost:8000/tryon \
  -F "person_image=@person.jpg" \
  -F "garment_image=@garment.jpg" \
  -F "seed=42"

Returns {"job_id": "...", "status": "pending"}.

Check Status

curl http://localhost:8000/tryon/{job_id}

Status: pending → processing → completed | failed

Download Result

curl -o result.png http://localhost:8000/result/{job_id}

Health Probes

Endpoint	Purpose
GET /health	Comprehensive status with GPU/disk info
GET /health/ready	Readiness probe (models loaded?)
GET /health/live	Liveness probe (process alive?)
GET /metrics	Prometheus metrics

Configuration

All settings use the MYFIT_ prefix:

Variable	Default	Description
MYFIT_HOST	127.0.0.1	Bind address
MYFIT_PORT	8000	Server port
MYFIT_DEVICE	cuda	cuda or cpu
MYFIT_PRECISION	float16	float16 or float32
MYFIT_RESULT_PATH	/tmp/myfit-results	Result storage
MYFIT_LOG_LEVEL	INFO	DEBUG, INFO, WARNING, ERROR
MYFIT_LOG_FORMAT	text	text or json
MYFIT_METRICS_ENABLED	true	Enable /metrics endpoint
MYFIT_TRACING_ENABLED	false	Enable OpenTelemetry
MYFIT_OTLP_ENDPOINT	—	OTLP exporter endpoint
MYFIT_FRONTEND_ENABLED	true	Enable web UI

Deployment

Kubernetes

kubectl apply -f k8s/

Includes Deployment with GPU resources, readiness/liveness probes, ConfigMap, Service, and Ingress.

NixOS

{
  inputs.myfit.url = "github:clairesrc/myfit";

  outputs = { self, nixpkgs, myfit, ... }: {
    nixosConfigurations.host = nixpkgs.lib.nixosSystem {
      modules = [
        myfit.nixosModules.default
        {
          services.myfit = {
            enable = true;
            host = "0.0.0.0";
            openFirewall = true;
          };
        }
      ];
    };
  };
}

Development

nix develop           # Enter dev shell
pytest tests/ -v      # Run tests
ruff check myfit/     # Lint
mypy myfit/           # Type check
pre-commit install    # Setup git hooks

CI/CD

GitHub Actions runs on every push:

• Lint: ruff check + format
• Type check: mypy
• Test: pytest with coverage
• Security: Trivy vulnerability scan
• Build: Docker image to GHCR (main branch)

License

MIT. The IDM-VTON model weights are licensed under CC BY-NC-SA 4.0 and are downloaded separately at runtime.