Community Guidelines

Content Guidelines Tips

Understanding what each layer catches can help you avoid unnecessary rejections and failed generations.

Layer 1 (Our Filter)

Layer 1 uses banned-word matching and the OpenAI Moderation API.

If your prompt is blocked at this layer, the task is rejected before generation and credits are fully refunded.

Common categories:

• harassment: Content that expresses, incites, or promotes harassing language towards any target. Input: text only.
• harassment/threatening: Harassment content that also includes violence or serious harm towards any target. Input: text only.
• hate: Content that expresses, incites, or promotes hate based on race, gender, ethnicity, religion, nationality, sexual orientation, disability status, or caste. Hateful content aimed at non-protected groups, such as chess players, is treated as harassment. Input: text only.
• hate/threatening: Hateful content that also includes violence or serious harm towards the targeted group based on race, gender, ethnicity, religion, nationality, sexual orientation, disability status, or caste. Input: text only.
• illicit: Content that gives advice or instruction on how to commit illicit acts. Input: text only.
• illicit/violent: Illicit content that also includes references to violence or procuring a weapon. Input: text only.
• self-harm: Content that promotes, encourages, or depicts acts of self-harm, such as suicide, cutting, and eating disorders. Input: text and images.
• self-harm/intent: Content where the speaker expresses that they are engaging or intend to engage in acts of self-harm. Input: text and images.
• self-harm/instructions: Content that encourages acts of self-harm or gives instructions on how to commit them. Input: text and images.
• sexual: Content meant to arouse sexual excitement, such as descriptions of sexual activity or promotion of sexual services, excluding sex education and wellness. Input: text and images.
• sexual/minors: Sexual content that includes an individual under 18 years old. Input: text only.
• violence: Content that depicts death, violence, or physical injury. Input: text and images.
• violence/graphic: Content that depicts death, violence, or physical injury in graphic detail. Input: text and images.

Layer 2 (ByteDance Content Review)

ByteDance's review system is significantly stricter than typical content guidelines enforcement. These categories commonly trigger rejections:

1. Real person depictions

Any content featuring real people is currently prohibited. This is stricter than many other video models. In some cases, fully synthetic faces may pass where real-person references do not.

2. Copyrighted or branded content

References to brands, franchises, studios, games, or trademarked properties are likely to be rejected. Generic descriptions work better than brand-name prompts.

3. Political content

Politically sensitive material is rejected.

4. Visual content review

Sometimes the prompt passes, but the generated frames still fail downstream review. In these cases, the request may fail during or after generation because the output visuals violate the provider's policy.

Typical Failure Patterns

Community guidelines rejection

You may see an error such as:

Your content violated community guidelines.

This usually means the prompt or generated visuals triggered provider community guidelines checks.

Server-side failure

You may also see a server-side failure message such as:

Video generation failed due to a server error. Please try again.

This is not a community-guidelines penalty. It indicates an upstream provider failure.

Credits and Refunds

When a Seedance 2.0 task fails on CyberBara, the credits consumed for that task are refunded automatically after the task is reconciled as failed.

The activity page also shows a refund note on failed tasks so users can confirm the credits were returned.

Practical Prompting Advice

• Avoid explicit sexual content.
• Avoid real-person references.
• Avoid brand names, franchises, and trademarked properties.
• Use generic visual descriptions instead of copyrighted character names.
• If a prompt fails repeatedly, simplify both the wording and the reference media.

Why This Exists

These community-guidelines checks are imposed by the upstream provider. We continue to improve the UX around failures, feedback, and refunds, but the underlying review rules are not fully under our control.