1.14.0 (2026-04-10)

Bug Fixes

• add FILES_STORAGE_PATH to Dockerfile ENV to prevent data loss (b575243)
• add shutdown timeout and improve health endpoint (986ad37)
• address code review findings before merge (caf65bc)
• correct PaddleOCR language codes for model download and OCR (e1ee571)
• force CPU mode in download_models.py for build-time compatibility (b4b59a7)
• handle paddlepaddle-gpu CUDA import at build time gracefully (0083a74)
• install cuda-compat stubs for build-time PaddlePaddle import (d31d665)
• load RealESRGAN pretrained weights for actual AI upscaling (fa9569c)
• revert to npx tsx in CMD for pnpm compatibility (e55253d)
• simplify smoke test to CPU-only imports for build-time compat (3481663)
• skip RealESRGAN import check on arm64 in smoke test (1e2ef52)
• split paddlepaddle-gpu and paddleocr installs, use --extra-index-url (74183e8)
• suppress ML library stdout noise in ocr.py and upscale.py (c0b419d)
• use PaddlePaddle GPU package index for CUDA wheels (dd9528f)
• use platform-specific mediapipe version for arm64 compatibility (7face19)

Features

• expand model pre-download with verification and smoke test (a9e3b96)
• simplify CI to single unified Docker build (b385a2e)
• simplify compose to single file, add log rotation (84f7057)
• unified Docker image with GPU auto-detection (6c3eb3b)