Seedance 2.0 Tips & Troubleshooting
Common issues, optimization strategies, and FAQ — everything you need to diagnose problems and get the most out of Seedance 2.0 video generation.
Why Negative Prompts Don't Work
If you're coming from Stable Diffusion or Midjourney, you might instinctively reach for negative prompts. Seedance 2.0 does not support negative prompts. There is no separate field for them, and including negative phrasing like "no blur" or "without distortion" in your prompt can actually cause the model to produce those exact unwanted effects — it latches onto the keyword "blur" and applies it.
The solution: always use positive, affirmative phrasing. Describe what you want to see, not what you want to avoid.
Wrong: "no blur, no distortion, without shaking, not grainy"Right: "sharp, crystal-clear detail, smooth stabilized camera, clean high-fidelity image"
Wrong: "avoid dark shadows, don't make it too saturated"Right: "soft even illumination, natural balanced color palette with muted tones"
Wrong: "no static camera, without zooming in too much"Right: "slow tracking shot following the subject at a steady medium distance"
Common Mistakes
These are the most frequent errors from both beginners and experienced prompt engineers. Each one has a straightforward fix.
Prompt too short (under 10 words) — Generic, random output. The model fills in the blanks with guesses. Fix: add detail — subject, motion, scene, camera, and style.
Prompt too long (over 150 words) — The model ignores parts of the prompt; elements compete for attention. Fix: keep to 30–100 words and cut redundant adjectives.
Too many style references — Visual chaos; conflicting aesthetics create an incoherent look. Fix: one strong style anchor plus lighting, remove competing styles.
Vague motion descriptions — Static or minimal movement; the model defaults to subtle idle motion. Fix: use power words like "dramatically", "forcefully", "explosively".
Inconsistent character descriptions — Character changes appearance between shots. Fix: use the exact same detailed description each time and anchor with @Image.
Wrong aspect ratio — Cropped or stretched output; composition feels off. Fix: match ratio to content type — 16:9 cinematic, 9:16 social, 1:1 square.
No camera direction — Random camera behavior; missed cinematic opportunities. Fix: specify shot type (close-up, wide) and movement (dolly, orbit, pan).
Overloading @references — Conflicting instructions from too many reference inputs. Fix: max 2–3 references per generation; keep them complementary.
Generic subjects — Boring, forgettable output with no visual interest. Fix: be specific — age, clothing, expression, distinguishing features.
Ignoring lighting — Flat, uninteresting visuals with no depth or mood. Fix: always include lighting direction and quality (golden hour, rim light, etc.).
Optimization Strategies
1. Start Simple, Then Layer. Begin with a basic 15–20 word prompt covering subject and motion. Once you get a result you like, add scene, camera, and style details incrementally. This helps you identify which element causes any unwanted changes.
2. Use the Prompt Formula. Always structure prompts as Subject + Motion + Scene + Camera + Style. This order mirrors how the model processes your input. Skipping elements forces the model to guess, which reduces output quality.
3. Check Your Word Count. The sweet spot is 30–100 words. Under 10 words yields generic results. Over 150 words causes the model to ignore parts of the prompt.
4. Anchor with One Strong Style. Pick a single style reference ("cinematic film", "anime cel-shaded", "film noir") and pair it with a specific lighting setup. Multiple competing styles create visual chaos and inconsistent output.
5. Power Words Control Intensity. Degree adverbs like "slowly", "dramatically", "gently", and "forcefully" directly control motion intensity and speed. Without them, the model defaults to medium-intensity motion.
6. Match Aspect Ratio to Content. Use 16:9 for cinematic and landscape shots, 9:16 for social media and vertical content, and 1:1 for product showcases. Wrong ratios lead to awkward cropping or wasted frame space.
7. Use @Image for Character Consistency. When creating multi-shot stories or sequences, reference the same character image with @Image in every generation. This ensures the model maintains appearance consistency across shots.
8. Specify Camera Always. Never leave camera direction to chance. Even "static wide shot, locked-off frame" is better than no camera instruction. The model generates more intentional and cinematic results when given explicit camera guidance.
9. Iterate, Don't Rewrite. When a result is close but not perfect, tweak one element at a time rather than rewriting the entire prompt. This helps you understand which changes improve or degrade the output.
10. Save Your Best Prompts. Keep a library of prompts that produced great results. They become reusable templates you can modify for new projects.
Iteration Techniques
Great results rarely come from a single generation. Follow this systematic workflow to refine your output progressively.
First Pass — Establish the Foundation. Write a simple prompt with subject and motion only (15–20 words). Generate and see what the model produces. This baseline tells you how the model interprets your core idea.
Review — Identify Gaps. Watch the output carefully. Is the motion too static? Is the scene wrong? Is the camera behavior random? Note exactly which elements need improvement before touching the prompt.
Adjust — Change One Thing. Modify only one element per iteration — add a camera direction, change the lighting, or refine the motion description. Changing multiple things makes it impossible to know what helped.
Second Pass — Validate. Regenerate with the single change. Compare side-by-side with the previous result. If the change helped, keep it and move to the next weak element. If it made things worse, revert and try a different approach.
Refine — Polish to Perfection. Once all major elements are working, add finishing touches — lighting nuances, style refinements, audio cues. These subtle additions elevate a good result to a great one.
Frequently Asked Questions
What's the best prompt length?The optimal range is 30–100 words. Under 10 words produces generic results because the model has to guess too many details. Over 150 words causes the model to deprioritize or ignore certain elements.
Can I use negative prompts?No. Seedance 2.0 does not support negative prompts. Including negative phrasing like "no blur" or "without distortion" can cause the model to produce those exact unwanted effects. Always use positive, affirmative descriptions of what you want to see.
How do I maintain character consistency across shots?Use the @Image reference tag to attach the same character reference image to every generation in a sequence. Also copy the exact same character description text across all prompts. Be specific: "a woman in her 30s with shoulder-length auburn hair, freckles, wearing a navy trench coat" is far more consistent than "a woman."
What resolution should I use?Seedance 2.0 supports up to 2K (1080p+). Use the highest resolution for final outputs. For rapid iteration and testing, lower resolutions generate faster. Always match the aspect ratio to your target platform: 16:9 for YouTube and cinematic, 9:16 for TikTok and Reels, 1:1 for Instagram.
How long does generation take?Typical generation times range from 20–60 seconds depending on resolution, duration, and server load. Seedance 2.0 is approximately 30% faster than v1.0. Shorter clips at 2K resolution take the most time. Real-time preview is available during generation.
Can I extend an existing video?Yes. Use Video-to-Video (V2V) mode by referencing your existing clip with @Video. For seamless extensions, use the last frame of your previous clip as an @Image reference along with a text prompt describing the continuation. This maintains visual continuity between segments.
How do I add sound to my video?Seedance 2.0 has built-in Auto Sound capability. Describe the audio you want in your prompt. For more control, use @Audio to reference a specific audio clip that sets rhythm, mood, and pacing. The model supports natural voice generation with lip-sync, environmental sound effects, and music scoring.
What's the @tag system?The @tag system allows you to reference multimodal inputs directly in your prompt. @Image attaches reference images for visual style, character appearance, or frame composition. @Video attaches reference videos for motion patterns, camera language, or as a base for transformation. @Audio attaches audio clips to drive rhythm, mood, or pacing. You can combine up to 9 images, 3 videos, and 3 audio files per generation.
Why does my output look different from the prompt?Common reasons: prompt too long (model deprioritizes later elements), conflicting instructions (mixing "calm serene" with "explosive action"), vague descriptions (too much freedom for the model), too many competing @references, or missing camera direction (model defaults to random behavior).
Can I use Seedance 2.0 commercially?Commercial usage rights depend on your plan on BigMotion. Generally, content generated with paid plans includes commercial usage rights, but always verify the specific terms for your use case, especially for client work or products for resale.
How do I create multi-shot stories?Write all your scene descriptions before generating anything. Use the same character description word-for-word across every prompt. Attach the same @Image character reference to each generation. Maintain consistent style and lighting descriptions. Use the last frame of each clip as an @Image reference for the next shot to ensure visual continuity. Vary camera angles between shots — wide, medium, close-up — for cinematic storytelling.
What languages does audio generation support?Seedance 2.0 supports English, Chinese, Japanese, Korean, Spanish, French, German, Portuguese, and more. For best lip-sync accuracy, write the spoken dialogue clearly in your prompt and specify the language. Environmental sounds, music, and sound effects are language-independent.
Prompt Glossary
T2V — Text-to-Video: generate video from a text prompt only.
I2V — Image-to-Video: animate a static image into a video clip.
V2V — Video-to-Video: transform, extend, or restyle an existing video.
Multimodal — accepting multiple input types (text, image, video, audio) simultaneously.
@Image — tag to attach a reference image for visual style, character, or composition.
@Video — tag to attach a reference video for motion patterns or camera language.
@Audio — tag to attach an audio clip to drive rhythm, mood, or pacing.
Dolly — camera physically moves toward (dolly in) or away from (dolly out) the subject.
Orbit — camera circles around the subject at a fixed distance and height.
Pan — camera rotates horizontally on a fixed axis, sweeping left or right.
Tilt — camera rotates vertically on a fixed axis, looking up or down.
Tracking — camera follows alongside a moving subject, maintaining steady distance.
Crane — camera rises or descends vertically, often revealing a scene from above.
Whip Pan — rapid horizontal camera rotation creating motion blur; used for transitions.
Rim Lighting — light behind the subject creating a glowing edge or silhouette outline.
Golden Hour — warm, soft sunlight shortly after sunrise or before sunset with long shadows.
Depth of Field — range of distance in focus; shallow DoF blurs the background for subject isolation.
Volumetric Light — visible light rays cutting through haze, fog, or dust (god rays).
Film Grain — analog texture added to digital video for a cinematic, organic feel.
Dappled Light — scattered light patterns created by filtering through leaves or obstacles.
Chromatic Aberration — color fringing at edges; a retro/VHS lens distortion effect.
Aspect Ratio — width-to-height proportion of the video frame (16:9, 9:16, 1:1, etc.).