does not follow

Welcome to AI video. For starters, most AI models have difficulty spelling words. Gemini is among the few exceptions. If you are asking for sound and not getting it, this may be a bug. But sound can also be added in post with a video editor. As for following instructions to the letter, it may (and often does) require multiple attempts to get the desired results. And regarding the video length, this can vary from model to model, I believe up to 12 seconds on some partner models offered by Adobe. Making shorter videos and editing them together is necessary for extended length.