imajin/tests

Test Infrastructure Overview

This directory contains the comprehensive testing infrastructure for the @image workspace.

Structure

@image/
├── test-all.sh                           # Workspace-level test runner
├── TESTING.md                            # Complete testing documentation
├── .github/workflows/test.yml            # CI/CD workflow
└── */tests/                              # Package-specific tests
    ├── conftest.py                       # Pytest fixtures
    ├── test_*.py                         # Unit tests
    └── integration/                      # GPU integration tests
        └── test_*_gpu.py

Quick Commands

Fast Unit Tests (No GPU)

./run test
# or
./test-all.sh

Duration: ~30-60s Coverage: API layer, validation, mocked pipeline

Full Test Suite (With GPU)

./run test --gpu
# or
./test-all.sh --gpu

Duration: ~5-10 minutes Coverage: Everything + real model execution

Test Types

Type	Location	Requirements	Duration
Unit Tests	tests/test_*.py	None	<1s each
API Tests	tests/test_*.py	Service running (some)	<1s each
Integration Tests	tests/integration/	GPU, CUDA, Models	10-30s each

Integration Test Coverage

image-pipeline (tests/integration/test_full_pipeline.py)

Tests: 12 GPU integration tests

• ✅ Photorealistic generation (SDXL)
• ✅ Anime generation (SD3.5)
• ✅ Different layouts (square, portrait, landscape, hero)
• ✅ All stages enabled (moderation, watermark, text overlay)
• ✅ Different inference steps (10, 20, 30)
• ✅ Different guidance scales (3.0, 7.5, 12.0)
• ✅ Deterministic generation (same seed = same output)
• ✅ Different output formats (PNG, WebP)
• ✅ GPU memory management
• ✅ Sequential generations
• ✅ CUDA diagnostics

image-generation/service (tests/integration/test_api_endpoints_gpu.py)

Tests: 10 GPU API integration tests

• ✅ POST /generate with photorealistic model
• ✅ POST /generate with anime model
• ✅ Hero layout generation (2048x1024)
• ✅ Text overlay integration
• ✅ Batch generation (multiple images)
• ✅ POST /warmup (model pre-loading)
• ✅ Model persistence between requests
• ✅ Performance with different steps
• ✅ GPU detection
• ✅ Model availability

Markers

Tests are marked with:

• @pytest.mark.gpu - Requires GPU (auto-skipped without CUDA)
• @pytest.mark.slow - Takes >5s to execute
• @pytest.mark.integration - Integration test (vs unit test)

Running Tests

All unit tests:

pytest -m "not gpu"

Only GPU tests:

pytest -m gpu

Specific package:

cd image-pipeline
pytest tests/integration/ -v -s

Single test:

pytest tests/integration/test_full_pipeline.py::TestFullPipeline::test_photorealistic_generation -v -s

Performance Expectations

RTX 3090, 20 steps:

• Square (1024x1024): 8-12s
• Portrait (768x1024): 6-10s
• Hero (2048x1024): 15-25s

First run:

• Model download: ~7GB, 10-30 min
• First generation: 20-40s (model loading)
• Subsequent: 8-15s (cached)

CI/CD

GitHub Actions (.github/workflows/test.yml):

• Runs unit tests automatically (no GPU)
• GPU tests skipped in CI (conftest.py auto-skips)
• Coverage reporting with Codecov

Documentation

See TESTING.md for:

• Complete testing strategy
• Writing tests guide
• Troubleshooting
• Best practices
• Commands reference

Architecture

Two-tier testing:

1. Unit/API Tests: Fast, mocked, CI-friendly

   • Test request validation
   • Test response structure
   • Test error handling
   • Mock expensive operations (GPU)

2. Integration Tests: Real execution, GPU-required

   • Test actual model loading
   • Test full 7-stage pipeline
   • Test GPU memory management
   • Test performance benchmarks

This allows fast development iteration (unit tests) while ensuring production readiness (integration tests).