Two Character Sheets, One Plate, One Prompt: A Complete Seedance 2.5 Video Workflow

Practical_Low29 · reddit · 2026-09-17

A Reddit user shared the complete workflow behind a Seedance 2.5 clip of two synthetic characters in an 8-second handheld take, built from just three reference images:

Two character sheets: GPT Image 2.5 four-panel layouts (front, three-quarter, full profile, rear) with matched lighting, consistent wardrobe, and visible pores. The author stresses both sheets must share identical lighting or characters look pasted together.
One location plate: a clean headshot plus a written scene (stone archway, bougainvillea, a woman by the fountain, the sea, point-and-shoot flash). The wardrobe line in the prompt keeps outfits consistent across shots.
One long video prompt: a reference map first (plate = frame zero, sets light and camera; sheets handle face/identity only), then an 8-second shot breakdown in time windows with in-lens effects, dialogue, and audio, plus an effects list, density map, and a hard-rules block longer than the description.

All models run via the Atlas Cloud ComfyUI node; all four prompts are in the comments. Core idea: grow the take from a still frame that already exists instead of letting the model pick a first frame.

Original post →

Two Character Sheets, One Plate, One Prompt: A Complete Seedance 2.5 Video Workflow

More from Multimodal