Jinyiwei Chase the Wanderer · Three Acts From Rooftops to Bamboo Scaffolds to the Bell Tower
This prompt splits one long chase into three acts marked ①②③ — rooftops, bamboo scaffold, bell tower — each swapping not just the backdrop but a whole prop mechanic: tiles shattering, ropes cut to topple the scaffold, the bell yanked off balance. Each act ends by stating exactly how the character enters the next (leaping across the street, grabbing a banner to swing toward the tower), concrete enough to actually cut together. Phrases like ‘keep the pursuer active,’ ‘both characters keep moving,’ and ‘constant repositioning’ repeatedly enforce continuous momentum so the model doesn’t freeze the chase into a standoff. Bullet-time is reserved for exactly one beat — the swinging bell — immediately followed by ‘return to full speed,’ so rationing the slow motion makes that one moment carry more weight.
This block isn’t a video prompt — it’s the creator’s workflow note explaining how the three-act chase actually got made: lock character sheets first so the wanderer and Jinyiwei stay visually consistent across clips, then prep separate visual references per location instead of sharing one set, so the rooftops, scaffold, and bell tower don’t bleed into each other’s look. Skipping a full storyboard in favor of key reference images plus detailed action prompts shows that sufficiently detailed text can substitute for a storyboard. Generating three separate clips and cutting the best shots together is the standard fix for long action sequences being unstable to generate in one go. Keeping only SFX and no score defers the music decision to post, so the model’s own music choice doesn’t lock in the final edit.
