Seedance 2.0 Troubleshooting Guide

💡 Solution:

• Use 3+ reference angles - Upload multiple angles of the same character (front, side, 3/4 view, close-up). This gives Seedance 2.0 more data to maintain consistency.
• Apply Identity-Lock - Enable the Identity-Lock feature (if available in your access method) to force the model to preserve character traits across shots.
• Reduce character descriptors - Simplify your Subject description. Fewer details mean less room for interpretation drift.
• Reference specific features - Call out distinctive, unchanging features like "tattoo on left cheek" or "silver hair with blue streak" to anchor consistency.

Woman @Image1 (for all shots, identity locked), short black hair, tattoos on left cheek. Walks through cyberpunk city street slowly...

💡 Solution:

• Rewrite Camera line first - Don't change Subject, Action, or Style. Only adjust the Camera parameters.
• Be specific with camera terminology - Use precise terms: "medium shot", "low-angle tracking", "slow dolly-in", "35mm lens" instead of vague words like "dynamic" or "cinematic".
• Test single parameter at a time - Generate with the new camera settings while keeping everything else identical. This isolates the variable.
• Reference shot size - If the framing worked in a previous generation, use the same shot size description as a template.

Original: A woman walking in rain...
Fixed: A woman walking in rain... [Camera: medium shot, low-angle tracking, 50mm lens]

💡 Solution:

• Specify movement speed explicitly - Add intensity modifiers: "slowly", "quickly", "gently", "violently", or "with large amplitude". This tells Seedance 2.0 how fast the motion should be.
• Use camera movement terms - Replace generic "dynamic" with specific camera actions: "handheld shake", "gimbal smooth", "slow pan left", or "tracking shot".
• Swap motion types - If "handheld" looks too shaky, switch to "gimbal" or "fixed tripod". If too smooth, add subtle handheld for realism.
• Describe physics realistically - Don't say "car turns". Say "The tires smoke as car drifts 90 degrees". Being specific about physical reactions helps.

Original: Car drives fast...
Fixed: Car drives, tires smoke as it drifts 90 degrees with large amplitude, camera tracks from side [Motion: dynamic, handheld shake]

💡 Solution:

• Use a single strong visual anchor - Replace multiple adjectives with one specific reference: "cinematic film noir", "Studio Ghibli inspired", "cyberpunk neon lighting". Strong anchors outperform vague style descriptions.
• Cite specific films or artists - Name a visual reference that matches your goal: "Blade Runner style", "Wong Kar-wai cinematography", or "Ghibli watercolor". This gives Seedance 2.0 a concrete style target.
• Be specific about color treatment - Instead of "colorful", use "warm golden hour lighting with orange and blue tones" or "neon blue and cyan cyberpunk palette".
• Use style keywords consistently - If using "anime", apply anime-specific terms throughout. Don't mix styles unless intentional.
• Add lighting description - Include light source and quality: "natural morning light from window", "neon sign reflection", "candlelight warm glow". Lighting anchors help color consistency.

Original: Cool style video...
Fixed: Cyberpunk city scene, neon blue and cyan lighting reflecting in puddles, Wong Kar-wai cinematography style [Style: Blade Runner, 1982]

💡 Solution:

• Keep hands visible but not prominent - Frame shots where hands are in the frame naturally but not the focus. This keeps them generated without emphasizing flaws.
• Use negative prompts for constraints - Add constraints like "no distorted hands", "normal human structure", "all fingers intact". While Seedance 2.0 doesn't support explicit negative prompts, you can phrase these as "with 5 fingers on each hand" to guide the model.
• Step back from close-ups - If close-up face shots are problematic, try medium or wide shots. Distance reduces the visibility of small facial features.
• Test at 5 seconds first - Generate shorter videos to check if the issue persists before using more credits on longer durations.
• Use multiple reference images - If using @Image1 for a character face, add additional angle references (@Image2, @Image3) showing the same face from different angles. This helps with consistency.

Original: Woman smiling at camera...
Fixed: Woman @Image1 (front view) smiling at camera, normal human structure, 5 fingers on each hand visible naturally in frame [Constraints: no distorted hands, all fingers intact]

💡 Solution:

• Use timestamp format - Break your prompt into timed segments: "[00-05s] Shot 1...", "[05-10s] Shot 2...". This guides Seedance 2.0 on precise timing for each scene transition.
• Add transition keywords - Include words like "fade to black", "cut to", "smooth transition" between shots. This helps create natural scene changes.
• Maintain consistent style across shots - Keep lighting, color treatment, and camera movement consistent throughout the multi-shot sequence unless a style change is intentional.
• Use "camera switch" keyword - Trigger shot changes explicitly with the "camera switch" command. This tells Seedance 2.0 when to change shots within a single generation.
• Generate shorter sequences first - Start with 2-3 shot sequences (10-15 seconds total) instead of long sequences. Easier to iterate and refine.
• Reference consistent character appearance - Use @Image1 for the main character in every shot to prevent drift. Don't change the @tag reference mid-sequence.

Original: Walking through city...
Fixed: [00-05s] Woman @Image1 enters city from left, [05-10s] she walks toward center, [10-15s] camera follows her movement, [15-20s] she exits frame right [camera switch]

💡 Solution:

• Test at 5 seconds first - Always start with shorter generations to establish pacing. A 5-second video uses fewer credits (~30) and gives quick feedback.
• Use speed modifiers consistently - Apply the same intensity level across all action verbs in your prompt. Don't mix "slowly" in one part and "quickly" in another.
• Break complex actions into beats - Instead of one long action description, split into sequential beats with timing: "Start: slow dolly-in. Then: gentle pan right for final 2 seconds."
• Match action to duration - For 15-second videos, describe 2-3 actions. For 5-second videos, describe 1 action. Longer videos need more content to fill the time naturally.
• Consider Creativity slider - Lower the Creativity/Consistency slider toward 60/40 favoring Consistency. This reduces unexpected pacing changes and maintains coherence.

Original: Walking fast...
Fixed: Woman walks slowly through city [Duration: 15s], picks up pace after 3 seconds [Pacing: consistent], continues walking toward exit