ByteDance’s 30-second single-pass AI video model launches today with 50 multimodal references and region-level editing, but ...