A named technique borrowing the VFX term 'plate' (background reference image): before using any reference image to generate video, edit out anything a video model could break or amplify — background clutter, stray/random objects, extra or garbled text. Video models are claimed to amplify these flaws rather than ignore them, so cleanup happens on the still reference before it ever reaches video generation. A preparatory discipline that sits upstream of the Three-Step AI Commercial Workflow (Assets → Setup → Generations)'s generations phase.
Beyond removing debris or logos, plate-cleaning also covers erasing unwanted structural elements from a location still before it's used as a generation reference — duplicate buildings, an extra row of market booths, or other extraneous detail the model would otherwise treat as real geography and try to preserve. The fix is manual: open the location image in a standard image editor and paint out the offending elements before feeding the still into blocking or scene generation. Example from the source: a location photo showing a duplicated building and an extra row of booths, cleaned up by hand-erasing the duplicates prior to use.
Из тем: Holding Continuity Across Independently Generated Shots