Free · copies cast, sets and prompts to your workspace · you pay only when you generate
About this template
Every recurring element gets its own reference sheet before any video is generated, and each video prompt points back to those sheets by number. The two survivors, Mia and Dave, were designed with GPT Image 2.5 Flare at 1K and 3:2 as model sheets on a cream background, with front and side bust portraits, full-body views and labeled color swatches for hair, eyes, skin and outfit. The plants follow the same approach: the mutated pea plant sheet shows a full view in soil, a side view and close-ups of its five-barrel head and glowing stem veins. Locations are made as empty plates with no people or creatures, such as the clapboard house front, the street and the living room window. In Seedance 2.5, each shot then lists its cast as @Image1, @Image2 and so on and asks the model to match each one exactly.
Each video prompt opens with the same global style block that asks for ungraded, untouched camera footage rather than a cinematic look. It specifies natural and practical light, low contrast, low saturation and a constantly shaking handheld camera that never stabilizes. Creatures are described as wet, veined living things that move in slow muscular pulses, while the zombies are framed as actors in prosthetics, which keeps the effects feeling practical. A performance block then describes emotion only through visible behavior such as breathing that moves the shoulders, blinking, swallowing and small flinches. The shots were generated in Seedance 2.5 at 480p and 16:9, with individual generations running from 18 to 30 seconds.
The canvas holds 16 reference image nodes plus a series of numbered Seedance 2.5 video generations, each covering one beat of the yard siege. Because the cast and locations live in separate sheets, you can replace Mia, Dave, a plant or the house with your own designs and leave the shot prompts mostly untouched. Rewrite the action lines for your story, keep the global style and performance blocks if you want the same found-footage feel, and regenerate. Finished clips land in the media pool and can be cut on the timeline and exported to mp4 or FCPXML. Inspired by a post by el.cine (@EHuanglu) on X.