Skip to main content

​
Short Answer

The “virtual-face whitelist” included with the Seedance 2.0 and 2.5 channels is an upstream capability enabled at the platform channel level. It does not need to be, and cannot be, manually enabled on your APIYI token or account. However, this does not mean that an image containing a face can always be submitted directly as a public URL or Base64 first frame or reference image. If an AI-generated virtual model is classified as “possibly containing a real person,” upload it to the Seedance 2.0 / 2.5 Asset Library, wait until its status becomes Active, obtain its asset://xxx ID, and use that asset ID in the video-generation task.
Channel access and trusted-asset status are different concepts
  • Virtual-face whitelist: Already available on the platform channel; no separate account or token activation is required
  • Asset ingestion: Registers a specific face image as a trusted asset and returns an asset:// ID that can be used for video generation

​
Do Seedance 2.0 and 2.5 Have the Same Face-Asset Rules?

For face-asset review, the current documentation gives Seedance 2.0 and 2.5 the same handling: neither generation can directly use a realistic real-person face as a first frame or reference image. This similarity is limited to face-asset review; it does not mean that the two model generations have identical capabilities or parameters.
  • An AI-generated realistic virtual person may still be classified as a possible real person
  • Virtual-person images should be ingested first and referenced with an asset://xxx ID
  • Real-person faces require liveness verification by the person involved and the corresponding authorized asset
SeeDance2 is a model-access group, not a bypass switch for face review. Seedance 2.5 and all 2.0-series models share the same SeeDance2 group (0.18x), so one token covers both generations with no extra group to select.

​
Why an AI-Generated Face Can Still Be Blocked

The upstream content-safety system assesses the image itself for faces or possible real-person content. It does not rely solely on a statement that the image was generated by AI. A highly realistic virtual model may therefore be classified as “possibly containing a real person,” and direct URL or Base64 input can still trigger anti-deepfake filtering. This normally does not mean that your APIYI account is missing a whitelist or that your token permissions must be changed. The correct solution is to ingest the image first so that it becomes a trusted asset.
Do not try to bypass face detection by repeatedly changing the URL, rehosting or compressing the image, or renaming the file. These operations do not change the face characteristics of the asset and cannot replace asset ingestion or real-person verification.

​
Correct Workflow for an AI Virtual Model

1

Sign in to the Asset Library

Open the icover.ai Asset Library, then register or sign in to your icover.ai account.
2

Upload the virtual portrait

Upload the AI-generated model under “Virtual Portrait Ingestion.” Supported formats include jpeg, png, webp, bmp, tiff, gif, and heic. The aspect ratio must be 0.4–2.5, each side must be 300–6000px, and each file must be under 30MB.
3

Wait for the asset to become active

Wait about ten-plus seconds until the status becomes Active, then copy the asset://xxx ID. An asset ID can be reused and does not need to be regenerated for every task.
4

Choose a generation method

In the icover.ai online generator, select Multimodal mode and set the reference-image type to Asset. For an API request, place asset://xxx in image_url.url.
5

Set the image role

Use role: "first_frame" when the asset is the first frame, or role: "reference_image" when it is a character or style reference. Refer to the asset as “Image 1” in the prompt.
For better character consistency, place a full-body frontal image and a neutral frontal face close-up of the same virtual character in one asset group.

​
Must I Use a Preset Asset or a Seedance-Generated Image?

No. All of the following sources can be used, but face assets should follow the appropriate workflow: The key requirement is not which model generated the image, but whether the face image has become an active trusted asset.

​
Can an AI Face Whitelist Be Enabled for My Account or for Seedance 2.5?

No separate activation is required, and there is no user-facing “AI face whitelist” switch on an account. The platform channel already includes the relevant upstream virtual-face capability. Your required action is asset ingestion, not a token-permission change. If a virtual portrait fails ingestion, or an active asset:// ID is still blocked, contact support and provide the following information so the asset state or upstream review result can be investigated:
  • Your APIYI account or registered email
  • The model name used
  • The asset ID (asset://xxx)
  • The video task ID and complete error message
  • A screenshot of the asset-ingestion status
When calling Seedance 2.5 or any 2.0-series model, confirm that the token has the SeeDance2 group selected. The group determines the model-access route and billing multiplier; it does not change face-review rules.
If the image actually contains a real person, or the system determines that real-person authorization is required, the upstream protection policy cannot be bypassed through ordinary virtual-person ingestion. Complete the performer’s liveness verification on the Real-Person Verification page and upload the asset to that performer’s dedicated verified asset group.

​
Common Questions About Seedance 2.5

No. For face-asset review, 2.5 and 2.0 use the same handling. Directly submitting a realistic real-person face may still be blocked by upstream content-safety controls.
No. SeeDance2 is the shared model-access group for 2.5 and the 2.0 series and controls routing and billing. Selecting it does not replace asset ingestion or real-person verification.

​
Directed-Access Notice for Seedance 2.0 / 2.5

Because the Seedance series uses direct official upstream resources and is subject to platform protection policies, it is currently intended primarily for selected customers with specific business needs. Its pages may therefore be hidden from the public documentation navigation, but they remain available through the following direct links:

Seedance 2.0 / 2.5 Overview

Review model capabilities, input restrictions, billing, and common questions.

Video Generation API

Review request formats for first-frame, first-and-last-frame, and multimodal reference generation.

Asset Library

Review virtual-person ingestion, real-person verification, asset requirements, and API endpoints.

Asset Reference Guide

Follow the complete upload, ingestion, Active polling, asset-reference, and video-download workflow.

​
Online Tools