The photo looks upright. The pixels your script reads can still be sideways. Before you blame the vision prompt, check one small piece of metadata: EXIF Orientation. Starter challenge: 1. Use a synthetic image with an obvious UP arrow. 2. Inspect its stored width, height, and orientation tag. 3. Compare the pixels your resize step actually receives. Does your preview agree with your photo viewer? The full AI Anonymous field note shows the missing preprocessing step, a runnable Pillow lab, eight authored orientation fixtures, and a smaller-preview adaptation. The article explains why the raw preview is deliberately saved without inherited metadata. We tested the local image exercise. No vision model, OCR accuracy, or performance claim was tested. Full walkthrough: https://jabsystems.io/ai-anonymous/pillow-orient-before-resize More practical field notes: https://jabsystems.io/ai-anonymous/#resources AI Anonymous by JABSystems, LLC #AIAnonymous #Pillow #Python #ComputerVision #AIWorkflows