Seedance 2.0 Best Practices

Use a clear structure: subject + action + environment + camera + style (+ sound if needed). Be specific; avoid vague words. For multi-shot, reference the same character or style in every prompt (e.g. “same character as @Image1”). Add constraints like “consistent outfit,” “natural anatomy,” “stable face” to reduce drift. For quality, some guides suggest ending with phrases like “4K, sharp details, cinematic, no blur, no flicker.”

Use high-resolution, well-lit images with a clear subject. Avoid cluttered or low-contrast references. For products, neutral backgrounds often work best. For character lock, one clear reference per character and refer to it in every shot. Keep file sizes and formats within platform limits.

Describe both subject motion and camera motion. Prefer simple, stable camera moves (dolly, truck, orbit, static). For action, “slow, smooth, continuous, natural” often yields more stable output; avoid overly complex multi-person or extreme motion unless you iterate.

Review each clip for character/style consistency and pacing. Regenerate or adjust prompts for shots that drift. Keep a small library of working prompts for your main use cases and refine from there.

Use one shared rubric before publishing any prompt example. Classify each daily run into the right page category (prompts, prompt-library, tutorial, or best-practices), then attach Input/Output evidence links when assets are ready.

Based on public tutorials and community reports, these are the most common Seedance 2.0 mistakes: (1) Forgetting @ references after uploading — upload your images/videos but never mention them with @Image1 or @Video1 in the prompt, so the model ignores them. (2) Tagging wrong media types — writing @Video1 when you uploaded an image, or vice versa. Double-check asset order. (3) Video extension duration — when extending a clip, the duration setting means the NEW part only, not total length. (4) Reference videos too long — keep reference clips under 15 seconds; longer clips may be truncated or cause errors. (5) Not generating multiple times — AI output is stochastic; generate 3-5 times and pick the best result. (6) Starting too complex — begin with a simple single-subject prompt, verify it works, then add references, camera, and audio incrementally. (7) Contradictory instructions — avoid putting both 'close-up' and 'wide shot' in the same prompt. (8) Ignoring file size limits — oversized files get rejected silently on some platforms; check the upload limits first.