Our validator checked files, not frames — the first external user taught us the difference

An open-source system called auto-xhs allows an AI agent to manage a Xiaohongshu account completely, from research to replying to comments. On the day it was released, user axiom3356 in Vietnam ran the system and discovered three visual issues that the validator, validate.py, missed. The first problem occurs during scene changes, where the screen goes nearly blank for around 2 seconds.

In the new scene, cards appear one by one, with nothing visible before the first one lands. The final video's bitrate was significantly low at 585 kb/s, which is far below the recommended 2 Mb/s for 1080p resolution. Additionally, two annotation cards in the same area with different delays can overlap each other.

The issue stems from the validator only checking the file's duration, resolution, word count, and silence gaps, but never examining individual frames. To address this, the system now includes checks that grab a frame every 0.5 seconds, compress it into a small grayscale image, and determine if it is nearly one solid color. If two consecutive frames exhibit this behavior, a warning is raised.

Similarly, a check was added to ensure the video bitrate remains above 2000 kb/s. These new checks are warned-only to avoid unnecessary alarm, while still informing users of potential problems. The issue remains open, awaiting a possible pull request from axiom3356, who discovered these issues and proposed the checks. Users are encouraged to share their experiences and create new playbooks within the scenarios/ directory of the auto-xhs repository.