How to Fix "Not Eligible" and "Visual Restriction" Errors on Seedance 2.0

You upload a reference image, enter your prompt, click generate — and instead of a video you get a red bar: "Not eligible" or "Visual restriction detected." No explanation of what triggered it, no hint about what to change.

You try a different image. Blocked again. You remove the face from the prompt but keep the image. Still blocked.

Seedance 2.0's moderation system runs three independent filters — on your text prompt, your reference images, and the generated output — and each platform that hosts the model configures these filters differently. A rejection on Dreamina does not mean the same image would be rejected on seedance2pro.io. A rejection today does not mean the same prompt would be rejected tomorrow.

This guide covers what "not eligible" and "visual restriction" actually mean in Seedance 2.0's filter system, which types of images and prompts trigger them, why faces are the most common cause, and what you can change — legally and safely — to get your request accepted.

If you are looking for a quick fix for region-based blocks rather than content-based blocks, see the Seedance 2.0 access guide. This article is about content eligibility — why your reference images and prompts are being rejected and what to do about it.

What "Not Eligible" Actually Means in Seedance 2.0

"Not eligible" is a content moderation rejection, not a technical error. It means one of Seedance 2.0's three filter layers decided your request violates its content policy and refused to process it.

Here is the critical distinction most users miss:

The first error is what this guide covers. If you are seeing the second, the fix is switching platforms — see the region access guide for that.

Seedance 2.0 operates three separate moderation checks:

1. Prompt filter — scans your text prompt before generation. Uses English keyword pattern matching.
2. Upload filter — runs computer vision on your reference images, specifically targeting photorealistic human faces and copyrighted visual elements.
3. Output filter — reviews the generated video after production and can reject it even if the prompt and references passed. This is the most frustrating block since credits are already consumed.

Most "not eligible" errors come from the upload filter rejecting a reference image — not from the text prompt. This means changing your wording alone often does not fix it.

What Is a "Visual Restriction" on Seedance 2.0?

A "visual restriction" is the platform's term for content the moderation system identified as potentially violating its image use policy. The restriction applies to the visual content of your reference image, not the text in your prompt.

The restriction flags images for one of three reasons:

1. Face detection. The filter identifies a human face in your reference image and blocks the generation. This is the most common trigger. Seedance 2.0's face upload filter specifically targets photorealistic human faces — illustrated, stylized, or AI-generated faces often pass where real photographs do not.

2. Copyrighted or trademarked visual content. The filter recognizes characters, logos, or design elements associated with known intellectual property. This includes franchise characters (even without naming them), brand logos, and distinctive visual motifs from major media properties. The filter detects visual concepts, not just text keywords — describing "a man in a red-and-gold armored suit" without mentioning Iron Man can still trigger a block.

3. NSFW or policy-violating content. Explicit material, realistic violence with weapons, and similar content are blocked across all platforms that host Seedance 2.0.

The "visual restriction" label is most commonly seen on Higgsfield's interface (the platform that originally popularized the term), but the same moderation system exists on Dreamina, Jimeng, and third-party platforms — just with different labels and thresholds.

Once you understand what causes a restriction, the next question is obvious: which images actually pass? The answer depends on the platform, but testing across all major hosts reveals a clear pattern.

Which Images Qualify as Eligible in Higgsfield / Seedance 2.0?

Image eligibility varies by platform, but testing across multiple platforms reveals a consistent pattern. The table below shows what typically passes vs. what gets rejected across most Seedance 2.0 platforms:

The grey area is real photographs of non-famous people — family photos, self-portraits, original headshots. These are sometimes blocked and sometimes accepted, depending on the platform and the quality of the image.

Testing shows that certain image characteristics increase the chance of rejection even for otherwise eligible content:

• Low resolution or heavily compressed images — the filter may flag these as suspicious
• Images with watermarks or text overlays — can be misread as brand content
• Images with multiple people — more faces = higher chance of detection
• Images with extreme lighting or heavy filters — the model may struggle to parse the content

That covers what generally passes and fails. But there is a pattern that confuses even experienced users: faces that pass sometimes and get rejected others — on the same platform, with the same image. Here is why.

Why Faces Work Sometimes but Not Others

This is the most confusing pattern users encounter. The same face image passes on one attempt and gets rejected the next — or works on one platform and fails on another.

Three factors explain this:

1. Platform-specific moderation thresholds.

Each platform that hosts Seedance 2.0 configures its own moderation strictness. ByteDance-owned platforms (Dreamina, Jimeng) apply the strictest face filters. Third-party platforms like seedance2pro.io, CyberBara, and TwoShot may apply different thresholds because they run the model on their own infrastructure with their own moderation configuration.

This is the most common reason for inconsistent face results: the same image rejected on Dreamina may pass on a third-party platform.

2. Face detection sensitivity depends on image quality and framing.

Seedance 2.0's upload filter uses computer vision to detect faces. If a face is partially occluded, shot from an unusual angle, small in the frame, or stylized away from photorealism, detection may fail — and the image passes. This creates the appearance of inconsistency when the actual variable is how clearly the face is visible.

Concretely:

• A full-face portrait shot in good lighting → almost always detected
• A side profile face with one eye obscured → sometimes detected
• A small face in a wide shot with other visual content → often passes
• A heavily stylized or filtered face → usually passes

3. Probabilistic scoring against a confidence threshold.

Seedance 2.0's moderation does not use a deterministic rule set. Instead, each filter layer assigns a confidence score between 0.0 and 1.0 to the content it evaluates — face likelihood, IP match probability, NSFW confidence — and compares it against a configurable threshold (typically 0.5, adjustable per platform). Content scoring 0.49 passes; 0.51 blocks.

The model runs on GPU hardware where floating-point operations introduce microscopic variability between invocations. Combined with input preprocessing variance (resolution downscaling, color normalization), borderline inputs can score 0.48 on one attempt and 0.52 on the next. This is why retrying the identical prompt sometimes passes on the third attempt — not a bug, but an emergent property of scoring-based moderation.

Face Filters in Seedance 2.0: What They Actually Check

Seedance 2.0's face filter is not a simple yes/no detection — it evaluates three separate dimensions:

The filter is notably stricter on English-language prompts than Chinese-language ones. ByteDance built its safety moderation primarily around English keyword lists and English-centric visual training data. Chinese prompts describing the same visual content often pass where their English equivalents do not.

This does not mean you can "bypass" the filter by switching languages — the upload filter operates on visual content, not text. But paired with a less aggressive prompt filter across all three layers, the overall rejection rate is measurably lower for non-English workflows.

The practical takeaway: if your prompt describes a face in English, the combined three-layer moderation is stricter than if your reference image alone carried the face information with a neutral non-English prompt.

Knowing how the filters work is half the battle. The other half is choosing where to run your generation — because the same filters behave very differently across platforms.

Which Platform to Use for Face Videos: Strict vs Permissive

Different platforms apply different face-filtering policies. If you need to generate videos that involve faces — even eligible ones like AI-generated characters — choosing the right platform is the single most impactful decision you can make.

ByteDance-Owned Platforms (Strictest Face Filter)

Dreamina (dreamina.capcut.com) and Jimeng (jimeng.jianying.com) apply ByteDance's default moderation, which is the strictest among all Seedance 2.0 platforms. Face detection on these platforms blocks most photorealistic face references regardless of whether the person is public, private, original, or AI-generated.

These platforms are best for:

• Landscape and environment videos
• Object and product animations
• Abstract and stylized content
• Content where face consistency is not required

Third-Party Platforms (Moderate to Loose Filter)

Third-party platforms apply their own moderation policies independent of ByteDance's defaults because they run the model on their own infrastructure. This is the single most important variable in whether a face reference passes or gets rejected.

API Providers (Most Flexible)

API access through fal.ai, Segmind, or Replicate offers the most flexibility for face usage. These platforms apply lighter moderation on API endpoints than on their web interfaces, and the moderation can sometimes be handled differently at the API layer. This makes API access the most viable path for developers who consistently need face-related generations.

For detailed API setup, see the Seedance 2.0 API guide.

How to Avoid "Not Eligible" Errors: A Practical Checklist

When your request gets rejected, work through this checklist in order. Each step addresses a specific filter layer.

Step 1: Check Your Reference Image (Covers ~70% of blocks)

Start with the upload filter — it rejects most requests.

• Does the image contain a visible human face? → Replace with an AI-generated character or stylized face
• Is the image a real photograph? → Switch to a stylized or illustrated version
• Does the image contain a brand logo, trademarked character, or recognizable IP? → Crop or replace
• Is the image resolution low or heavily compressed? → Use a clean high-resolution image
• Does the image have watermarks or text overlays? → Remove them
• Is the image the only reference? → Try replacing it with a prompt-only generation first

Step 2: Check Your Prompt (Covers ~20% of blocks)

If the image passes but the prompt fails, the text filter is responsible.

• Does your prompt name a real person (celebrity, politician, public figure)? → Describe the character role instead
• Does your prompt name a copyrighted character or franchise? → Describe the visual qualities without naming
• Does your prompt contain violence keywords (shoot, kill, weapon)? → Use cinematic equivalents (muzzle flash, impact, tactical gear)
• Does your prompt use explicit or NSFW language? → Remove or replace with neutral descriptions
• Is your prompt in English? → Consider a neutral non-English description if you are facing repeated text filter blocks

Step 3: Try on a Different Platform (Covers remaining blocks)

If image and prompt both look clean but you still get rejected, the platform's moderation configuration is likely the issue.

• Try the same image + prompt on seedance2pro.io
• Try without face references on Dreamina
• Try through API access (fal.ai, Segmind) for reduced moderation
• Retry the same request 2–3 times — borderline cases sometimes pass on rerun

Step 4: Confirm the Block Layer

Not all "not eligible" errors come from the upload filter. Use this diagnostic to identify which layer is rejecting your request:

What to Do When Nothing Works: Two Honest Options

Some content genuinely cannot be generated on Seedance 2.0 under any platform's policies. When you have exhausted the checklist above and still get rejected, you have two honest options — and one clear warning about what not to try:

Option 1: Change the Creative Approach

If the specific image or prompt you are using keeps getting blocked, instead of trying to force it through, reframe the creative brief:

• Replace a real-person reference with an AI-generated character that achieves the same visual goal
• Replace a copyrighted character reference with an original character using similar but non-infringing visual traits
• Replace a photorealistic face reference with a stylized portrait that provides character guidance without triggering face detection
• Generate the character outside Seedance (in Midjourney, DALL·E, or Stable Diffusion), then use that AI-generated output — which faces no likeness restrictions — as the reference

Option 2: Use a Different Model

For content that consistently fails across all Seedance 2.0 platforms, consider whether a different AI video model would accept the same input with fewer restrictions. Kling 3.0, for example, has a less aggressive moderation setup for action sequences and face references.

What Not to Do

Attempting to bypass moderation by cropping faces, using heavy filters, or exploiting prompt ambiguities is counterproductive. These attempts often fail at the output filter layer — meaning you waste credits on a generation that gets rejected after processing. Worse, repeated moderation violations on ByteDance-controlled platforms can lead to account restrictions.

Still have questions? Here are the most common ones users ask after reading through this guide.

FAQ

What does "not eligible" mean on Seedance 2.0?

It is a content moderation rejection. One of Seedance 2.0's three filters — prompt, upload, or output — detected content that violates the platform's content policy. The most common cause is a photorealistic human face in a reference image.

What is a visual restriction on Seedance 2.0?

A visual restriction is the moderation system's rejection of a reference image based on its visual content. It typically means the image contains a detected face, copyrighted visual element, or policy-violating content. The term is most commonly used on Higgsfield's platform.

Which images are eligible in Higgsfield / Seedance 2.0?

AI-generated characters, landscapes, objects, stylized/illustrated faces, and abstract visuals typically pass. Real photographs of people, public figures, copyrighted characters, brand logos, and NSFW content are typically blocked. Eligibility varies by platform — third-party platforms like seedance2pro.io are more permissive than ByteDance-owned ones.

Why can I use faces on some Seedance 2.0 platforms but not others?

Each platform configures its own moderation policies independently. ByteDance-owned platforms (Dreamina, Jimeng) apply the strictest face and IP filters. Third-party platforms (seedance2pro.io, CyberBara) may use different thresholds. API providers (fal.ai, Segmind) typically apply the least restriction.

Can I generate videos with faces on Seedance 2.0?

Yes, but the platform matters. Third-party platforms and API access are more likely to accept face references — especially AI-generated or stylized faces. ByteDance-owned platforms like Dreamina and Jimeng have a near-complete ban on photorealistic face references.

What triggers the face filter in Seedance 2.0?

The face filter checks three dimensions: likeness to known public figures, photorealism score, and context classification. A face that is photorealistic, resembles a known person, and appears in a sensitive context is almost certain to be blocked.

How do I fix "not eligible" on Seedance 2.0?

Work through the four-step checklist: (1) Check and replace the reference image, (2) Check and revise the prompt, (3) Try a different platform, (4) Confirm which filter layer rejected the request. For most users, step 1 (replacing a real face with a stylized or AI-generated one) solves the problem.

Where to run Seedance 2 Pro with faces?

Third-party platforms like seedance2pro.io offer a more permissive face policy than ByteDance-owned platforms. For developers, API access through fal.ai or Segmind provides the most flexibility. The model is identical across all platforms — only the moderation configuration differs.

Bottom Line

"Not eligible" and "visual restriction" errors on Seedance 2.0 are frustrating because they give you no explanation of what triggered them. But the underlying system is predictable once you understand the three filter layers and how each platform configures them.

Here is the practical summary in one sentence: the upload filter rejects most requests, photorealistic faces are the most common trigger, and switching to a third-party platform fixes most blocks without changing your creative direction.

• If your reference image contains a face: Replace it with a stylized or AI-generated version, or switch to a platform with a more permissive face policy like seedance2pro.io.
• If your prompt contains names or explicit keywords: Describe roles and visual qualities instead of naming specific people or franchises.
• If the image and prompt look clean but it still fails: The platform's moderation policy is the bottleneck. Try a different platform or API access.
• If nothing works across multiple platforms: The content itself may not be compatible with Seedance 2.0's current moderation system. Consider a different creative approach or model.

Your next step (takes 2 minutes): Open your current reference image. If it contains a visible face, replace it with an AI-generated character from Midjourney or Stable Diffusion — keep the pose and composition the same. Upload the AI-generated version on seedance2pro.io with your original prompt unchanged. That single swap resolves the majority of "not eligible" errors without changing your creative direction. If you need a step-by-step walkthrough of the full Seedance 2.0 workflow, see the Complete Guide.