In the prompt's reference block, each asset gets a description of its personality and quality, but no description of how it looks. The attached images already carry the look, so repeating it in words adds a second, possibly conflicting source for the same information.
This fits the same logic as minimal-video-prompt-references-hold-identity: let the references hold identity and use words for what images cannot show. For where the reference block sits in the whole prompt, see Five-block video prompt structure.