Use an image-to-video workflow (Seedream still → Seedance animation) instead of text-to-video so Seedance can match a reference image and keep the character/dish consistent across shots.
Only send “keeper” Seedream frames to Seedance (filter out wrong faces/objects first) to avoid paying to animate incorrect outputs and cut your Seedance render cost.
Write two prompts per shot (a “dish/character file” build prompt + a per-shot action prompt) and copy the build prompt word-for-word across all shots to maintain consistent visuals when animating in Seedance.
Prompt Seedance with a mid-air “throw apart” action and specify the desired landing effect (e.g., “lands the steam”) to get dynamic food disassembly/reassembly motion with visible steam.
