8af9f12544
PIL's strict load was raising OSError on images missing the trailing end-of-stream marker (e.g. '6 bytes not processed' on a JPEG without its FF D9 EOI), failing the entire ML task for an image the model could otherwise score fine. Set ImageFile.LOAD_TRUNCATED_IMAGES = True in both ML modules so a minutely-corrupt tail doesn't block tagging or embedding. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>