In this Seedance 2.0 beginner prompting guide, I walk you through how I actually structure prompts so the model follows your instructions “to the bone.” I start with the core prompt format I use most: opening scene (what we see first), the action (what happens), the shot/details (4K, lighting, environment), and optionally audio. I also explain why I almost always recommend reference-based generation (same credit cost as text-to-video) because it keeps characters consistent and reduces distortions—especially if you’re doing a specific character style. I show how to use multiple references (up to 9 images, plus videos/audio) and why clean, uncluttered character references on plain backgrounds work best. Then I test more advanced workflows: using a ChatGPT prompt generator (Noble Goose) to auto-format prompts with image tags and mapping, and a “per-second” prompt format that tells Seedance exactly what should happen at each moment for more precise results. Finally, I cover practical loopholes and safety tips: how some violent prompts can get flagged and why translating to Chinese Mandarin can sometimes pass stricter Western filters, why mixing languages can cause weird accent flips, and how to prevent annoying auto-subtitles or random music by adding “no subtitles” and “no music” at the end of your prompt.