How we test claims
ImgifyTools publishes what the deployed stack actually does. Marketing language is restored only after readiness checks, automated behavioral tests, and a manual golden-image review—not from a green /ready flag alone.
Privacy baseline
Per-tool facts live in lib/tool-privacy.ts. Example for background remover: Uploads are processed on ImgifyTools servers for your session, are not retained permanently, and are not used to train AI models. Maximum upload size is 10 MB.
Readiness flags
- serving_ready — rembg + OpenCV guided refine available for cutouts
- upscale_ready — Real-ESRGAN stack + x2/x4 weights; otherwise LANCZOS resize
- face_ready — GFPGAN weight present
- object_inpaint_ready — LaMa available; otherwise OpenCV inpaint
- cartoonizer_ready — OpenCV filters (always; not neural style transfer)
Behavioral tests
Backend suite tests/test_ai_behavioral.py checks task-specific outcomes: meaningful alpha masks, correct upscale dimensions (and non-LANCZOS difference when ready), face-region vs background change, masked vs unmasked object removal, and cartoon pixel change. Converters assert format and dimensions.
Manual golden review
Before enabling NEXT_PUBLIC_AI_MARKETING_ENABLED, a human reviews side-by-side golden images for face enhancer, upscaler, background remover, and object remover. Automated green alone does not unlock AI marketing copy.