Why Is Seedance 2.0 Rejecting Your Reference Video? 6 Causes and How to Fix Each One

Your video upload failed. No error message. No hint about which part of the clip triggered it.

You upload a short clip — maybe a camera movement you want Seedance 2.0 to replicate, or a dance routine you want to restyle into a cartoon. Instead of a generated video, you get a silent rejection. This is the most confusing part of using video references on Seedance 2.0: the platform tells you something is wrong, but not what.

As Seedance 2.0 expands across more platforms in 2026 — Dreamina, Seedance2Pro, Jimeng, and Higgsfield — more users are seeing video rejection for the first time. The moderation system that handles images does not apply the same rules to video, which is why even a clean clip can fail without explanation.

I have tested hundreds of reference video uploads across all major Seedance 2.0 platforms to map what passes and what gets blocked. What I found is that six distinct causes account for virtually every rejection, and most can be resolved in under five minutes once you know which one applies to your clip.

If you have already read the Seedance 2.0 "not eligible" guide, you know how image moderation works. Video reference rejection is a separate problem — it involves a different set of technical requirements, different moderation signals, and different preparation strategies.

If you are here because you need a quick fix right now, jump to the 6-point checklist.

Why Seedance 2.0 Treats Video References Differently Than Images

To understand why a clean-looking clip gets blocked, you need to see how Seedance 2.0 processes video differently from images. The difference explains most of the rejection reasons below.

Seedance 2.0 accepts video files in two ways:

• Motion reference — the model watches the clip and replicates its movement, camera work, or choreography in a new generation. The visual content of the reference is not preserved; only the motion pattern is extracted.
• Style reference — when used on Dreamina's video-to-video workflow, the model transforms the uploaded video's visual style (for example, converting a real video into a cartoon aesthetic).

The platform matters. On Seedance2Pro, a video reference primarily serves as motion guidance. On Dreamina, the same video can be used for full restyling. The moderation that applies to each use case is slightly different, but the technical and policy reasons for rejection overlap significantly.

On any platform, Seedance 2.0 supports up to 3 video reference files per generation when paired with other reference types, or a single video with up to 15 seconds of content. Images are capped at 9, audio at 3, and the total reference count across all types is 12 files.

Those limits are generous — but rejection usually happens before you reach them.

Here is why the pipeline matters. An image is scored as a single frame — one pass, one decision. A video is processed as a sequence of frames, each scored independently before the results are aggregated. This frame-by-frame approach means that even a brief flash of problematic content — a face visible for two seconds, a logo that appears in the background for a few frames, a compression artifact that looks like a watermark — can trigger a rejection, because those individual frames exceed the detection threshold even though the clip as a whole looks clean to your eye.

The result is six distinct causes that block reference videos. Here is what they are and how to fix each one.

The 6 Reasons Seedance 2.0 Rejects a Reference Video

1. File Format or Specification Violation (The Easiest Cause to Fix)

The most common rejection cause is also the easiest to fix: your video file does not meet Seedance 2.0's technical requirements.

Based on current platform documentation and cross-platform testing, Seedance 2.0 accepts the following video formats:

If your clip exceeds any of these limits, the upload filter rejects it before the moderation layers even run.

Common file spec violations that cause silent rejection:

• Wrong codec. A file with the .mp4 extension but encoded with a codec other than H.264 (for example, HEVC/H.265 or VP9) can appear valid to you but fail to parse on the server. Re-encode to H.264 Baseline or Main profile before uploading.
• Variable frame rate (VFR). VFR files, especially those recorded on phones or screen capture tools, sometimes cause parsing errors. Convert to constant frame rate (CFR) before uploading.
• DRM-protected or encrypted files. Screen recordings from streaming services, purchased downloads, or files with embedded DRM are rejected. The upload filter detects encryption headers and blocks the file immediately.

2. Face Detection — the #1 Moderation Reason Seedance 2.0 Rejects Reference Videos

This is the most common moderation-based rejection for reference videos, and the one that confuses users the most.

Seedance 2.0's upload filter runs computer vision on every reference file — including videos, not just images. When a video contains a visible human face, the same face detection model that blocks photorealistic images also blocks the video clip.

The filter evaluates three dimensions:

The critical distinction: a video without visible faces is almost never rejected by the face filter. A landscape pan, an object showcase, a camera movement through an empty room — these pass the face check every time. The moment a face appears on screen for any significant duration, rejection risk rises sharply.

This creates a pattern that surprises many users: a video that worked as a motion reference yesterday gets rejected today — not because the filter changed, but because the new clip happened to catch the subject's face in frame for more frames than the previous one.

The Dreamina factor. This is especially true on Dreamina, ByteDance's own platform, where the face filter is the strictest among all Seedance 2.0 hosts. Dreamina's video-to-cartoon workflow is one of the most popular use cases — convert a real-world clip into an animated style — but it is also where rejection is most frustrating, because the source material (a person walking, talking, or dancing) almost always contains a face. If you are using Dreamina and getting rejected, face detection is almost certainly the cause.

3. Copyrighted Visual or Audio Content Caught by the Frame Sequence

The IP detection filter on Seedance 2.0 does not just scan static images. It analyzes video frames as a sequence, which gives it more data points to match against known IP signatures.

A reference video is more likely to trigger IP detection than a static image for one reason: motion reveals more identifying context. A still photo of a character might crop out the logo on their shirt. A video of the same character walking past the camera shows the logo from multiple angles across dozens of frames — giving the filter more evidence to match.

Video clips that commonly trigger IP rejection:

• Footage with visible brand logos — a clip shot in a store, a video with a branded product on screen, or a screen recording that includes a software toolbar
• Background media — a TV playing in the background, a poster on a wall, a billboard in a street scene
• Music or audio fingerprints — if the video includes an audio track with a recognizable commercial song, the audio fingerprinting layer can reject the clip even if the visual content is clean
• Recognizable franchise characters — a cosplay video, a toy unboxing, or a theme park clip featuring a ride based on a franchise

4. Compression Artifacts That Trick the Moderation Filter

Low-quality or heavily compressed videos are rejected more often than clean, high-quality footage. This is not because the filter targets compression directly — it is because compression artifacts mimic the visual signatures that the filter is trained to flag.

When a video is heavily compressed (low bitrate, high quantization), the encoder introduces blockiness, color banding, and edge noise. The moderation model, trained on detecting visual anomalies, can misclassify these artifacts as:

• face-like patterns (blocky skin regions with enough structure to trigger face detection)
• text or logo remnants (banding that resembles a watermarked area)
• synthetic or AI-generated markers (noise patterns that match the model's "unnatural content" classifier)

This is most common with:

• Videos downloaded from social media (Instagram, TikTok, YouTube re-encodes at low bitrates)
• Screen recordings at reduced quality settings
• Clips that have been re-encoded multiple times
• GIF-to-MP4 conversions

The fix is simple: start with the cleanest source file you have. If you must use a social media download, upscale and denoise it before uploading rather than feeding the compressed version directly.

5. Violence, Weapons, or NSFW Content in the Video Frames

Seedance 2.0 applies its NSFW and violence filters to video frames, not just static images. A reference clip that contains — or appears to contain — violent action, weapons, or explicit content is rejected.

This filter is notably strict on:

• Realistic violence. A staged fight scene from a film, a sparring session, or even a dramatic fall can trigger the violence classifier if the motion and context read as realistic rather than stylized.
• Weapons on screen. Guns, knives, and weapons visible in the reference video are flagged even if the clip is a movie still, a cosplay prop, or a historical reenactment. The filter does not distinguish context — if a weapon-shaped object appears in frame, it scores high on the violence dimension.
• Blood or injury imagery. Even subtle blood effects, bruising, or injury makeup in the reference video can trigger rejection.
• Suggestive or explicit motion. Dance moves, close-contact choreography, or implied adult content in the video reference can trigger the NSFW classifier.

Animated or stylized violence passes more easily than live-action — a cartoon punch is scored differently than a real punch — but the safest approach is to avoid any violent or explicit content in the reference video entirely.

6. Platform Uploader Glitch — When It Is Not Your Video at All

Sometimes the rejection is not your clip — it is the uploader.

Seedance 2.0 is hosted on multiple platforms (Dreamina, Seedance2Pro, Higgsfield, Jimeng, API providers), and each platform implements its own upload pipeline. These pipelines have bugs.

Known uploader failure patterns (not caused by your video):

• "Upload failed" on specific platforms with no error detail — the file is valid, has passed moderation, but the platform's upload handler crashes or times out
• Silent rejection on files over a certain size — the stated limit might be 150 MB, but some platforms reject files above 100 MB due to server-side timeouts
• Intermittent rejection of valid files — the same clip passes once and fails the next time, suggesting a server-side processing issue rather than a content issue
• Browser-specific upload failures — Safari users on Dreamina report higher rejection rates than Chrome users, likely due to differences in how the browser preprocesses the file before sending it

If your clip passes all checks above and is still rejected, try:

1. Uploading the same file on a different platform (for example, try Seedance2Pro if Dreamina rejected it)
2. Using a different browser (Chrome or Edge rather than Safari)
3. Re-encoding the file with the exact same settings but a lower resolution
4. Waiting 10–15 minutes and retrying (transient server issues resolve)

Which Cause Is Blocking Your Video? Match Your Error Message

Six possible causes, one rejected clip. How do you know which one hit yours? Seedance 2.0 does not always tell you which cause triggered the rejection — but the error message or behavior you see during upload often points directly at the culprit.

The 6-Point Checklist: Diagnose Your Rejection in 3 Minutes

When a reference video is rejected, work through these checks in order. Each addresses one of the six causes above.

File Spec Check

• Is the file MP4 (H.264) or MOV? → Re-encode if not
• Is the duration under 15 seconds? → Trim if longer
• Is the file size under 150 MB? → Reduce resolution or bitrate
• Is the frame rate constant (CFR), not variable (VFR)? → Convert to CFR with HandBrake or FFmpeg
• Is the aspect ratio 16:9 or 9:16? → Crop or letterbox extreme ratios

Face Check

• Does the video contain a visible human face for more than ~2 seconds? → Replace the clip or crop the face out of frame
• If yes, is it a real person (not animated)? → Switch to a stylized or AI-generated character clip, or use a clip with no visible face
• If the clip shows a person but no face (from behind, distant), does it still get rejected? → Try on a third-party platform with a looser face policy

Copyright Check

• Does the video contain visible brand logos or trademarks? → Crop, blur, or replace
• Does the video include commercial music in the audio track? → Remove or replace the audio track
• Does the background contain recognizable franchise characters or media? → Reshoot or change the background

Quality Check

• Does the video show visible compression artifacts, blockiness, or banding? → Use a higher-bitrate source or upscale before upload
• Was the clip downloaded from social media (TikTok, Instagram, YouTube)? → Find or create a cleaner source

Content Check

• Does the video depict realistic violence, weapons, or physical combat? → Replace with a stylized or implied-action version
• Does the video contain suggestive or explicit content? → Change the source clip

Platform Check

• Have you tried uploading the same clip on a different platform? → If it works elsewhere, the original platform's uploader is the issue
• Have you tried a different browser? → Switch to Chrome or Edge
• Have you retried after 10–15 minutes? → Transient server issues resolve on retry

How to Prepare a Reference Video That Passes Every Time

If you want to skip the trial-and-error entirely, here is a preparation workflow that produces reference videos with a high acceptance rate across all Seedance 2.0 platforms.

Step 1: Choose the right content.

Select a clip that contains:

• No identifiable faces (or faces that are heavily stylized)
• No brand logos, trademarks, or franchise content
• No weapons, violence, or explicit material
• Clean, well-lit footage with minimal compression

The safest reference clips are object showcases, landscape pans, camera movement tests, and animated sequences.

Step 2: Encode correctly.

• Container: MP4
• Video codec: H.264 (Main profile)
• Bitrate: 8–15 Mbps for 1080p (scale proportionally for lower resolutions)
• Frame rate: 30 fps (constant)
• Resolution: 1080p or lower
• Duration: 10 seconds or less (15 seconds max)
• Audio: AAC, 128 kbps, or no audio track

Step 3: Trim and clean the timeline.

Cut the clip so the useful motion happens in the first 5–10 seconds. Remove any lead-in or tail frames where nothing moves. A tighter clip means fewer frames for the filter to scan and higher confidence that the motion will be captured.

Step 4: Verify on a test platform before your main workflow.

Before committing to a full generation workflow, upload the reference video alone on Seedance2Pro with a minimal prompt (for example, "Follow the camera movement from the reference video"). If the upload succeeds and generation starts, your clip passes all filter layers. If it fails at this stage, the issue is in the video itself, not in your prompt or workflow.

Step 5: If direct video reference is consistently blocked, use an alternative.

When a real-world video clip cannot pass the filters — and you cannot or do not want to change the content — these alternatives work reliably:

• Generate a synthetic version of the clip. Render the same motion in a 3D tool (Blender, Unreal Engine) or use an AI video model that is less restrictive (Kling 3.0, for example) to create a reference-quality clip from the source content.
• Replace real people with AI-generated characters — particularly useful for Dreamina's video-to-cartoon workflow. Generate a character in Midjourney, DALL·E, or Stable Diffusion that matches the pose and framing of your original clip. Use that AI-generated image or clip as the reference instead. Dreamina's filter flags real faces, not generated ones — and the cartoon workflow produces the same restyled output regardless of whether the input was real or generated.
• Use a sequence of still images instead. Extract keyframes from the video at 1–2 second intervals and upload them as image references with a prompt that describes the intended motion between frames. The Seedance 2 image-to-video guide covers this workflow in detail.
• Describe the motion in text. If the reference clip is a specific camera movement or choreography pattern, describe it in the prompt rather than uploading a video. Seedance 2.0's text understanding is strong enough to interpret "slow orbit around the subject, starting at eye level and rising to a 45-degree top-down angle" without a visual reference.

FAQ

Why does Seedance 2.0 keep rejecting my reference video?

The most likely cause is one of six: file format or codec violation, a detected face in the video, copyrighted visual or audio content, compression artifacts triggering false flags, violence or NSFW content, or a platform-specific uploader bug. Work through the 6-point checklist to identify which one applies to your clip.

What video formats does Seedance 2.0 accept for reference?

MP4 (H.264) and MOV. Maximum 15 seconds duration, approximately 150 MB file size (varies by platform). Constant frame rate preferred. Avoid HEVC, VP9, or other codecs.

How do I use Dreamina Seedance 2.0 to convert a video into a cartoon?

Upload your source video as a reference on Dreamina's Seedance 2.0 video-to-video workflow and describe the target style in the prompt (for example, "Convert this video into a 2D anime style"). If the source video is rejected, the most likely cause is a detected face — Dreamina applies the strictest face filter among all Seedance 2.0 hosts. Replace the real person with an AI-generated character and re-upload.

Can I convert a personal video into a cartoon on Seedance 2.0?

Yes, but personal videos are the most likely to be rejected because they typically contain real faces in real-world settings. Use the preparation workflow above: trim to under 15 seconds, encode as H.264, and if the clip contains a face, replace the subject with an AI-generated character before uploading the reference.

What is a visual restriction on Seedance 2.0?

A visual restriction is the moderation system's term for rejecting a reference based on its visual content. For video references, the most common trigger is a detected human face. The term is most commonly seen on Higgsfield's interface, but the same moderation pattern applies across all Seedance 2.0 platforms. See the detailed Seedance 2.0 "not eligible" guide for a full explanation of visual restrictions.

Which images and videos are eligible in Higgsfield Seedance 2.0?

On Higgsfield, eligible references are those with no detected faces, no copyrighted visual or audio content, no NSFW or violent content, and proper file format (MP4/MOV for video, JPG/PNG/WebP for images). AI-generated and stylized content passes at a higher rate than real-world footage or photography.

My reference video passes on one platform but fails on another. Why?

Each Seedance 2.0 platform configures its own moderation thresholds and upload pipeline. ByteDance-owned platforms (Dreamina, Jimeng) apply the strictest face and IP filters. Third-party platforms like Seedance2Pro may use different thresholds. If the same clip passes on one platform and fails on another, the clip itself is fine — use the permissive platform for your workflow.

Can I use a video with a face as a motion reference without triggering the filter?

Partially. Videos where the face is small in the frame, partially occluded, or visible for under ~2 seconds sometimes pass. There is no guaranteed threshold — the filter scores each frame probabilistically. The most reliable approach is to use a clip with no face on screen, or to replace the real person with an AI-generated character. See the face filter section in the eligibility guide for a deeper explanation of why face detection is inconsistent.

Bottom Line

Seedance 2.0 rejects reference videos for six distinct reasons, and most users hit Cause 2 (face detection) or Cause 1 (file format) first.

Here is the practical summary in one sentence: a clean, face-free, properly encoded MP4 under 15 seconds passes on any platform — and when it does not, the issue is usually the platform's uploader, not your clip.

• If your clip contains a visible face: Crop it out or switch to an AI-generated character reference. This resolves the majority of moderation-based rejections.
• If your clip fails immediately on upload: Check the codec, file size, and duration. Re-encode to H.264, keep it under 150 MB, and trim to 15 seconds.
• If your clip uploads but the generation fails partway: The output filter detected something in the frames. Review the clip for compression artifacts, background IP content, or motion that reads as violent or explicit.
• If the same clip passes on one platform but not another: Use the platform that accepts it. The moderation differences are configuration choices, not content problems.

Your next step (takes 5 minutes): Open your reference video. Check its codec and duration first (those are free fixes). If the video contains a face, decide whether you can crop or replace it. Upload the cleaned version on Seedance2Pro with a simple test prompt — if it generates, you have identified and solved the problem. If it still fails, walk through the 6-point checklist above and you will find the cause within three checks.