Some in-scene details sit outside a video model's reliable rendering range — small text on a prop, a newspaper headline, anything that needs to read correctly at a glance. Rather than trust the video generator to render the detail live, generate the prop as its own standalone image in a separate pass (pre-processed as needed, e.g. pre-blurred to match depth of field), then drop the finished asset directly into the video prompt as a reference.
This is a distinct step from Clean Your Plate (Reference Image Prep for Video Generation) (which subtracts unwanted elements from an existing plate) — here the workflow adds a purpose-built prop the main model would otherwise get wrong. It pairs well with the Higgsfield @-Element Asset System, where an approved pre-rendered prop can be registered as its own reusable @-element.
Example from the source: a newspaper headline needed for a shot was generated separately and pre-blurred to match the shot's depth of field, then inserted as the reference asset instead of asking the video model to write and render the text itself.
Из тем: Holding Continuity Across Independently Generated Shots