Do not embed auto playing multimedia with either audio or visual components.
All multimedia objects require the person to opt in by manually selecting a play button or other engagement element that clearly defines the behavior being initiated. All controls must be properly labeled and ARIA accurately represent its current state.
Captions are required for all videos with at least 99% accuracy.
Auto captioning services alone do not meet this requirement. Captions support all users, but are specifically necessary for people who are hard of hearing.
Audio description is required where visuals are used to deliver meaning.
To support people who are visually impaired, audio description must accurately represent the content visually presented for the person to have the ability to derive the same meaning from the video content. If the visual content is dense, create a second version of the video with extended audio description that would otherwise be disruptive to other people.
All audio must include transcription using HTML format.
This supports all people, including those who prefer to read, but especially those who cannot easily hear. With the transcription in HTML, the content also gets an SEO boost, so it can more easily be found through search engines.
Never use video as background
The use of video as background with content overlaid has been proven to be an unfortunate design trend that causes motion sickness and in most cases lack controls.