Full prompt workflow for a 24s Cleopatra ad with Seedance 2.5

socialwithaayan · x · 2026-11-01

Creator @socialwithaayan breaks down the full pipeline behind a 24-second "Egyptian queen" AI spot:

Step 3, reference images first: Claude generates 4 reference images (queen, black panthers, robot, throne hall) to keep characters consistent across clips; full prompts are shared, including identity anchoring from an attached face photo and 35mm film texture.
Step 4, video: with Seedance 2.5, upload the 4 images in order and use @Image 1/2/4 reference syntax, then write shot-by-shot prompts at 0.7s per beat (nobleman offering a diamond, the bored queen flicking it away, a panther snapping it mid-air...).
Step 5: assemble clips in order, add music, cut on the beat.

More from Multimodal