Free · copies cast, sets and prompts to your workspace · you pay only when you generate
About this template
Each character and prop gets its own reference sheet, and every video prompt names those sheets by role. The template holds 6 reference image nodes made with GPT Image 2.5 Flare at 2K, 16:9: HARPER the art student, WONG the old man, BIGFOOT the creature, the CART, and two empty street plates. In the video prompts each reference is tagged with its name, the model is told to take only the face, hair, clothing or fur, and to disregard the sheet layout and labels. The cart prompt also states there is only one cart, so it never duplicates mid-shot. Even in the later extension, the same tagged references come back, so the cast looks the same in every shot.
The main shot was written as one staged scene with clear geography, then generated with Seedance 2.5 at 1080p, 16:9, 30 seconds. Two plates of the same steep San Francisco street were generated as empty photographs with soft daylight, low contrast and muted color: one facing uphill, the other taken from the top of the hill with the bay below. The prompt places HARPER at the bottom on the bay side, so the uphill push and the downhill view stay readable. The story summary does the rest: the old man struggles with the cart of cans, the student draws a gentle creature, throws the sketchbook, and the drawing appears on the rails to push the cart up the hill.
The ending is an extension shot generated with Seedance 2.5 Edit at 720p, 16:9, 15 seconds. It reuses the same tagged references and adds a new beat: after the student leaves, the creature mischievously sends the cart rolling back down and laughs while the old man runs after it. The prompt asks for fast handheld, raw-footage camerawork and frames WONG facing the creature with the street falling away to the bay behind it. The old man, out of breath, removes his cap and pats the creature's arm, with dialogue in English with a light Cantonese accent. This extend-then-cut approach lets you add a punchline without regenerating the whole scene.