Quick answer
If a Seedance 2 real-face upload is rejected, check the portrait policy before changing the prompt. BytePlus's current documentation restricts direct image and video references containing real human faces for Seedance 2.0 and 2.5. It provides specific portrait-asset workflows instead. Permission to use someone's photo does not automatically enable those workflows on your chosen service. BytePlus LAS input requirements
This guide explains supported routes, upload checks, and what to do after a rejection. Provider, account, region, model, and mode can affect availability.
What “Seedance real faces” can mean
Three different questions often get grouped together:
- Realistic appearance: Can a video contain a believable fictional human character?
- Real-person identity: Can a reference reproduce a particular real person's likeness?
- Character consistency: Can the subject keep the same appearance throughout a shot or across clips?
A realistic-looking character does not establish permission to reproduce someone's identity. Likewise, a prompt asking for “the same face” describes a continuity goal; it does not authorize the reference or change an upload restriction.
Decide whether your project needs a specific real person before preparing assets. A fictional presenter, product-only demonstration, or illustrated character may serve the creative brief without using that person's likeness.
Choose a supported route before uploading
Use this decision table to identify the next step. It is a planning checklist, not a promise that a particular provider will accept an asset.
| Your situation | What to check | Next step |
|---|---|---|
| A selfie or another person's photo was rejected | Exact error and the selected service's portrait policy | Confirm whether an authorized real-person workflow is available |
| You have the subject's permission | Verification, onboarding, account and regional access | Follow the provider's documented authorization process |
| You have an AI-generated portrait | Whether that exact asset qualifies for a supported portrait route | Check its origin and eligibility before reuse |
| You need a fictional presenter | Available digital characters and permitted original assets | Choose a supported character with no intended real-person identity |
| An accepted character changes appearance | Reference roles, action complexity and output review | Address consistency after confirming eligibility |
For the last situation, the character reference sheet guide covers preparation and visual review. Those techniques address drift; they do not establish upload eligibility.
The portrait routes documented by BytePlus
The official ModelArk portrait guide describes three routes for Seedance 2.0 and 2.5:
- Trusted model outputs. Eligible original face-containing outputs from specified models can be reused within the same ModelArk account for 30 days. Cross-platform, cross-account, edited, or expired outputs do not qualify. Output moderation still applies.
- Preset digital characters. ModelArk provides character assets identified by asset IDs for projects that need a realistic-style person without specifying a real individual.
- Authorized real-person assets. The person completes real-person verification and authorization. Assets are then added to the library and referenced using their asset IDs.
These are ModelArk workflows. They do not establish that our Seedance 2.5 service or another third-party interface exposes the same features. This site's public upload controls were verified; an authorized real-person onboarding flow was not verified for this service.
Before buying credits for a portrait project, ask the provider whether the required route is enabled for your account and region, how verification works, and how approved assets are selected. Keep identity verification inside the provider's official process.
What to do when a real-face upload is rejected
1. Keep the exact error
Record the message, selected model, task mode, file type, and whether the failure happened during upload or generation. Save the job ID if one exists. A size error, unsupported format, account-access problem, and content rejection call for different responses.
Avoid treating every failed upload as a face restriction. If the message is unclear, ask support which requirement failed before spending credits on repeated attempts.
2. Separate consent from technical access
Having the subject's approval is an important starting point. Taking the photo yourself also does not settle whether the service accepts it as a portrait reference.
ByteDance's Seedance 2.0 launch announcement says real-human portrait references require identity verification or prior legal authorization. Read that alongside the current provider workflow; it does not promise that every interface has a direct-upload option.
3. Check the documented asset path
For a service offering a portrait library, confirm how approved assets are referenced and whether processing has completed. A local filename or an ordinary upload label should not be assumed to be a verified portrait-library asset.
BytePlus LAS separately describes its material library as an allowlisted feature, using library IDs for authorized real-person references. Access to a general video API therefore should not be assumed to include that feature. LAS asset input documentation
4. Stop retries aimed at defeating the restriction
If the rejection is policy-based, use a supported asset route or revise the project. Do not alter a real person's face to slip through detection. A prompt saying “fictional” does not turn a real-person reference into an original character.
Upload limits depend on the surface
File limits and portrait eligibility are separate checks. A correctly sized reference can still fail the content requirements.
On October 4, the public Media to Video workspace opened with Seedance 2.5 Lite selected. It displayed limits of 9 images, 3 videos, and 3 audio files, with file-size hints of 30 MB, 50 MB, and 15 MB, respectively. Standard reference was enabled; Video edit and Video extend were disabled for that selection.
By comparison, the official LAS documentation lists 1–9 images for Seedance 2.0 multimodal references and 1–30 for Seedance 2.5, with each image below 30 MB. These are LAS specifications, not this website's upload allowance. LAS input requirements
Recheck the selected model and enabled mode before submission. For assigning roles to accepted inputs, read the Media to Video reference guide.
Use an original fictional character when identity is unnecessary
Start from an original character design rather than a modified photograph of a real person. Choose a permitted asset type or an available provider character library. An AI portrait generated on another platform should not be assumed to qualify as a trusted ModelArk output.
Give the character a simple role in the story, such as demonstrating a workspace or introducing a product. Keep the first shot easy to review. Check the returned clip for unwanted resemblance, changing features, and unintended dialogue or claims before building a longer sequence.
For supported product images or original illustrations, Seedance 2.5 Image to Video provides a starting workspace. Review the selected model, current input rules, and displayed credit cost first.
FAQ
Does Seedance 2.5 automatically allow real faces that Seedance 2.0 rejects?
No. The current BytePlus restriction covers both versions. Check the provider's supported portrait route rather than assuming a version upgrade changes eligibility.
Can I use my own face if I consent?
Check whether your provider offers the required verification and asset workflow. Your consent alone does not establish technical access or guarantee acceptance.
Does successful generation mean I can publish any depiction?
Review the subject's approval for the actual script, edits, audience, and intended use before publishing. Do not imply that a real person endorsed a product or made a statement they did not approve. Keep approval records with the project.
What should I check about portrait privacy?
Read the chosen service's data-processing, storage, retention, and deletion information before uploading. The Seedance 2.5 Privacy Policy states that submitted prompts and images are sent to third-party AI providers for processing. Ask support about portrait-specific handling if the policy does not answer your question.
Sources
Checked October 4, 2026: BytePlus ModelArk portrait asset guide · BytePlus LAS input requirements · ByteDance Seedance 2.0 launch announcement.
