"The Request Failed Because… May Contain Sensitive Information": Seedance & Seedream Errors Explained
TL;DR: These are ByteDance's content-safety errors. Seedance (video) and Seedream (image) share the same set of 21 codes. Each code tells you what was checked: your input or the model's output, and whether it was text, an image, a video or audio. Its suffix tells you why: general sensitive content, copyright (.PolicyViolation), a real person (.PrivacyInformation) or counterfeit documents (.DeepFake). Fix the part the code names. Resending an unchanged input usually gets the same answer.
Where These Errors Come From
ByteDance serves Seedance and Seedream through ModelArk on BytePlus and through Ark on Volcano Engine in China. Both platforms run the same content-safety checks and return the same codes.[1] BytePlus describes two layers: an input filter that screens what you send, and output moderation that screens what the model makes.[2]
A blocked request returns HTTP 400 with a code and a message. Every message in this family starts the same way:
The request failed because the input image may contain sensitive information.Request ID: {id}.
Apps and APIs built on these models usually pass the error along, sometimes wrapped in their own text or replaced with a shorter message. The missing space before "Request ID" is in ByteDance's own message, so it's normal to see it.
How to Read the Code
Each code is built from three parts. Take InputImageSensitiveContentDetected.PrivacyInformation as an example:
- Input or Output: whether the check ran on what you sent or on what the model made. Here, the input.
- Text, Image, Video or Audio: which part was flagged. Here, an image.
- The suffix after the dot: why it was flagged. Here, a real person.
There are four possible reasons:
- No suffix: general sensitive content, such as sexual, violent or otherwise restricted material.
.PolicyViolation: the content "may be related to copyright restrictions"..PrivacyInformation: the content "may contain real person"..DeepFake: the output "may contain counterfeit documents or credentials".
The stage matters most. An Input code means nothing was generated: change the file or prompt it names. An Output code means your request passed and the result was blocked: change what you asked for.
Most messages end with a Request ID. Keep it if you plan to contact support.
Every Seedance and Seedream Content-Safety Code
ByteDance's error table lists these codes for the whole platform.[1] The "Seen with" column is based on what each model takes in and puts out: only Seedance accepts video and audio references or makes video, and only Seedream makes still images.
Prompt checks
| Code | Message ends with | Seen with |
|---|---|---|
| SensitiveContentDetected | …input text may contain sensitive information. | Seedance, Seedream |
| SensitiveContentDetected.SevereViolation | …input text may contain severe violation information. | Seedance, Seedream |
| SensitiveContentDetected.Violence | …input text may contain violence information. | Seedance, Seedream |
| InputTextSensitiveContentDetected | …input text may contain sensitive information. | Seedance, Seedream |
| InputTextSensitiveContentDetected.PolicyViolation | …input text may be related to copyright restrictions. | Seedance, Seedream |
Reference file checks
| Code | Message ends with | Seen with |
|---|---|---|
| InputImageSensitiveContentDetected | …input image may contain sensitive information. | Seedance, Seedream |
| InputVideoSensitiveContentDetected | …input video may contain sensitive information. | Seedance |
| InputAudioSensitiveContentDetected | …input audio may contain sensitive information. | Seedance |
| InputImageSensitiveContentDetected.PolicyViolation | …input image may be related to copyright restrictions. | Seedance, Seedream |
| InputVideoSensitiveContentDetected.PolicyViolation | …input video may be related to copyright restrictions. | Seedance |
| InputAudioSensitiveContentDetected.PolicyViolation | …input audio may be related to copyright restrictions. | Seedance |
| InputImageSensitiveContentDetected.PrivacyInformation | …input image may contain real person. | Seedance 2.0 and 2.5 |
| InputVideoSensitiveContentDetected.PrivacyInformation | …input video may contain real person. | Seedance 2.0 and 2.5 |
Output checks
| Code | Message ends with | Seen with |
|---|---|---|
| OutputImageSensitiveContentDetected | …output image may contain sensitive information. | Seedream |
| OutputVideoSensitiveContentDetected | …output video may contain sensitive information. | Seedance |
| OutputAudioSensitiveContentDetected | …output audio may contain sensitive information. | Seedance (with audio) |
| OutputTextSensitiveContentDetected | …output may contain sensitive information. | Text models only |
| OutputImageSensitiveContentDetected.PolicyViolation | …output image may be related to copyright restrictions. | Seedream |
| OutputVideoSensitiveContentDetected.PolicyViolation | …output video may be related to copyright restrictions. | Seedance |
| OutputAudioSensitiveContentDetected.PolicyViolation | …output audio may be related to copyright restrictions. | Seedance (with audio) |
| OutputImageSensitiveContentDetected.DeepFake | …output image may contain counterfeit documents or credentials. | Seedream |
Seedance and Seedream never return text, so you shouldn't see OutputTextSensitiveContentDetected with them. It covers the language models on the same platform.
Generate with Seedance 2.5 on Imagenter AI
Seedance 2.5 runs with ByteDance's own safety checks. If a generation is blocked, your credits come back automatically.
"…input text may contain sensitive information": Fix the Prompt
The request failed because the input text may contain sensitive information.
Codes: SensitiveContentDetected and InputTextSensitiveContentDetected. The prompt itself was refused, so nothing was generated.
- Find the restricted part. Sexual content, graphic violence, drugs, weapons aimed at people, self-harm, hate, and real public figures are the usual triggers.
- Watch for double meanings. Harmless words can read as risky out of context: "shoot" in a photo shoot, or "naked" in "naked eye".
- Rewrite only the risky sentence. Describe what the camera sees in plain words, and keep the rest of the prompt.
- Don't resend it unchanged. The same prompt usually gets the same answer.
Two variants are more specific:
.SevereViolation("may contain severe violation information") means the filter put your prompt in its most serious category. Rewording rarely helps. Drop the concept..Violence("may contain violence information") is about how violent the description is. Cut the graphic detail, such as blood, wounds or weapons aimed at someone. An action scene can stay; a gore close-up can't.
"…input image may contain sensitive information": Replace the File
The request failed because the input image may contain sensitive information.
The request failed because the input video may contain sensitive information.
The request failed because the input audio may contain sensitive information.
A reference file was refused before generation started. Rewriting the prompt won't help; the file is the problem.
- Check the whole frame, not just the subject: posters, screens, tattoos, printed text and signs in the background all count.
- If the flagged detail isn't something you need, such as a poster on the wall or a print on a shirt, use a version of the file without it.
- For audio references, check the words as well as the sound. Spoken lines and lyrics are screened too.
- Test files one at a time when you send several. The code doesn't say which one failed.
"…may contain real person": Seedance's Real-Face Rule
The request failed because the input image may contain real person.
The request failed because the input video may contain real person.
Codes: InputImageSensitiveContentDetected.PrivacyInformation and InputVideoSensitiveContentDetected.PrivacyInformation.
BytePlus built "technical controls to restrict making videos from images or videos that contain real faces" into Seedance, to stop impersonation and unauthorized use of people's likeness.[3] The check runs on any photo or video of a real person, including one of yourself. It says "may" because it's a classifier: a very realistic painting or render can trip it too.
There is an official route, but it isn't for most users. BytePlus lets vetted business clients "complete real-person verification and likeness authorization through the ModelArk console", and ModelArk also offers a library of over 10,000 virtual human assets.[3] Apps built on the API usually can't offer either.
What works for most people:
- Start from a clearly non-photographic character: an illustration, anime, clay or stylized 3D design.
- Describe the person in words and use text-to-video, with no face reference at all.
- Keep references for everything else: the setting, the outfit, a product, a camera move.
Don't try to disguise a real face to get it through. The rule exists to stop impersonation, and getting around it can break the terms of the app you're using.
"…may be related to copyright restrictions": the PolicyViolation Codes
The request failed because the input image may be related to copyright restrictions. Request ID: {id}.
The request failed because the output video may be related to copyright restrictions. Request ID: {id}.
All seven .PolicyViolation codes are about copyright. ByteDance runs protection designed to prevent the generation of unauthorized intellectual property.[3]
- Input codes mean a file or the prompt looks like protected material: a film or TV still, a game or anime character, a brand logo, album art, or a copyrighted song in an audio reference.
- Output codes mean your inputs passed, but the result came out too close to existing work. This usually happens when the prompt names a character, franchise or brand.
To fix it, remove names and signature details, and describe an original design instead: its own outfit, colors and silhouette. Use your own footage and music, or material you have a license for. If a project needs a specific character or brand, the route is permission from the rights holder, not a reworded prompt.
"…output video may contain sensitive information": the Result Was Blocked
The request failed because the output image may contain sensitive information.
The request failed because the output video may contain sensitive information.
The request failed because the output audio may contain sensitive information.
Your prompt and files passed. The model made something, and the output check blocked it.
- Generation is random, so the same request can produce a result that passes. If the idea is clearly harmless, one retry is reasonable.
- If it fails twice, change the prompt. The model is reading something in it as more explicit or violent than you meant. Common causes are revealing clothing, close body framing, injuries and blood.
- For Seedance with audio, check any dialogue or lyrics you wrote into the prompt. The generated audio is checked separately.
"…may contain counterfeit documents or credentials": the DeepFake Code
The request failed because the output image may contain counterfeit documents or credentials.
Code: OutputImageSensitiveContentDetected.DeepFake. Seedream made an image that looks like a forged document or credential, such as an ID card, passport, driver's license, bank card or certificate.
There's no prompt fix for a realistic copy of a real document. If a film or app mockup needs one on screen, design an obviously fictional document with an invented issuer, and expect realistic look-alikes of real ones to be refused.
Messages That Mention ContentSecurityDetection
Volcano Engine also runs a separate risk-identification service. Requests it screens out fail with a different message:
The request could not be processed because the input text includes sensitive content that violates ContentSecurityDetection.
The codes are InputTextRiskDetection, InputImageRiskDetection, OutputTextRiskDetection and OutputImageRiskDetection.[1] Their messages carry Label and SubLabel fields that name the category. The fixes are the same as above: change the input for an Input code, change the prompt for an Output code.
ContentSecurityDetectionError ("Internal error…") is different. It means the risk check itself failed to run, not that your content was flagged. Try again later.
When the Error Looks Wrong
False positives happen, and every message says "may". To find the trigger:
- Test the prompt alone, without any reference files.
- Add files back one at a time until the error returns.
- Change one thing per attempt, so you know which change fixed it.
If a harmless request still fails, send the Request ID to the support team of the app or API you use. Only they can raise it with ByteDance.
You can't switch the checks off. BytePlus says "Some platform-level safety policies cannot be disabled."[2]
Whether you're charged depends on the platform; ByteDance's error code docs don't say. On Imagenter AI, a failed Seedance or Seedream generation refunds its credits automatically.
Blocked Generations Don't Cost You Credits
Use Seedance 2.5 for video and Seedream 5.5 for images on one credit balance. Failed generations are refunded automatically, and every request follows our Acceptable Use Policy.
Free trial credits • No subscriptions • Pay-as-you-go
FAQ: Seedance and Seedream Content Errors
What does "The request failed because the input image may contain sensitive information" mean?
ByteDance's input filter refused an image you uploaded, before anything was generated. The code is InputImageSensitiveContentDetected. Replace the image, or use a version without the flagged detail. Rewriting the prompt won't change the result.
What is OutputVideoSensitiveContentDetected in Seedance?
Seedance made the video, then the output check blocked it. Your prompt and files had already passed. If the idea is clearly harmless, retry once. If it fails again, reword the parts of the prompt that could read as explicit or violent.
Why does Seedance say my image "may contain real person"?
Seedance 2.0 and 2.5 restrict making videos from photos or videos of real people, to prevent impersonation. Any photorealistic face can trigger the check, including a photo of yourself. Use a stylized character instead, or describe the person in the prompt and leave the face reference out.
What does PolicyViolation mean in a Seedance or Seedream error?
Copyright. Every code that ends in .PolicyViolation says the input or output "may be related to copyright restrictions". Remove characters, franchises, brands and logos from your prompt and reference files.
What does OutputImageSensitiveContentDetected.DeepFake mean?
Seedream's output looked like a counterfeit document or credential, such as an ID card, passport or certificate. Realistic copies of real documents are refused, and there is no prompt change that makes them allowed.
Can I turn off the Seedance or Seedream content filter?
No. BytePlus says some platform-level safety policies cannot be disabled, so apps that run these models can't switch them off for you either.
Is this the same error as "Green net check rejected"?
No. Green Net is Alibaba Cloud's filter for Wan and Qwen models, and it returns DataInspectionFailed. The codes on this page come from ByteDance's ModelArk and Volcano Engine Ark, which serve Seedance and Seedream.
Am I charged when Seedance or Seedream blocks a request?
That depends on the platform you use; ByteDance's error code docs don't say. On Imagenter AI, a failed Seedance or Seedream generation refunds its credits automatically.
References
- Error codes — Volcano Engine Ark documentation.
- Why was my request rejected for sensitive or non-compliant content, and how do I reduce false rejections? — BytePlus Help Center.
- Dreamina Seedance 2.5 is Now Available on BytePlus — BytePlus, Aug 6, 2026.
