Seedance 2.0 Asset Library (Character-Consistent Video)

Use the icover.ai Asset Library open API (an APIYI product) to ingest images and get asset:// IDs, then generate character-consistent videos through the APIYI gateway. Supports zero-code web operation and REST API batch integration.

icover.ai is a sub-product of APIYI — an online tool for testing AI video generation. This asset library and its API help developers and customers ship “character-consistent video” features.

Why an asset library

When generating character-consistent videos, Seedance 2.0 does not accept reference images containing human faces directly (anti-deepfake filtering). You must first ingest the image as a trusted asset, get an asset://xxx asset ID, and reference that ID in the video-generation request.

This service handles the ingest for you: just upload an image → get an asset ID → generate the video. There are two ways to use it, and the data is fully shared between them:

Method	Best for	Requirements
Web UI	Everyone, zero code	Just sign up and log in
API	Developers who need programmatic batch integration	An “Asset Library KEY”

Shared data: for the same account, assets uploaded via the web and via the API live in the same library — API-created assets show up in the web asset list / archive and in the video generator’s reference-image picker, and web-created assets can be listed via the API.

The asset library follows your icover.ai account, not the KEY: all Asset Library KEYs created under the same account are equivalent and access the same library — a KEY is just a call credential and provides no isolation. Assets are isolated per account; you can only see and operate on your own. For isolation and multi-team sharing options, see the FAQ below.

Two different keys — do not mix them up:

• Asset Library KEY (created on icover.ai, sk-...): only for the asset-library endpoints on this page (upload / ingest / query / delete).
• APIYI SeeDance2 token (created on api.apiyi.com, sk-..., with the SeeDance2 group enabled): only for the video-generation endpoint.

Method 1: Web UI (recommended for beginners)

1. Sign up and log in

   Open the icover.ai asset library page and sign up / log in.

2. Upload and ingest

   On the “Virtual Avatar Ingest” tab: select images (multi-select supported) → click “Upload & Ingest”.

   • The asset group is optional; the system automatically uses your default group. Create a group first if you want to organize assets per character.
   • Image requirements: jpeg / png / webp / bmp / tiff / gif / heic; aspect ratio 0.4–2.5; side length 300–6000px; under 30MB each.
3. Copy the asset ID

   Wait about ten-plus seconds; once the status turns “Active”, copy the asset://xxx asset ID.

4. Generate the video

   Go to the icover.ai video generator: switch to “Multimodal” mode, set the reference image type to “Asset”, select your asset, and refer to the character as “Image 1” in the prompt.

Real-person assets (web version): three steps on the “Real-Person Verification” tab — 1) click “Generate verification link”, have the actor scan the QR code / open the link on their phone, log in to their Volcengine account, and complete the liveness verification; 2) click “Query verification result” to get the actor’s dedicated real-person asset group; 3) select that group and upload assets (images / video / audio) — they pass a face-consistency check and you get the asset:// IDs. The same actor in different styling reuses the same group, no re-verification needed.

The asset library web UI: the ingest tab for manual uploads, the asset list / archive tab for browsing assets and copying asset:// IDs, and the real-person verification tab for verifying and uploading real-face assets

Character-consistency tip: put a full-body frontal shot and a neutral frontal face close-up of the same character into one asset group for the best results.

Method 2: API (developers)

Step 0: Create an Asset Library KEY

After logging in to icover.ai, go to Settings → Asset Library KEY (icover.ai/en/settings/apikeys) and create a KEY in the sk-... format.

Settings → Asset Library KEY: click the create button and copy the generated sk-... key (note this is separate from the APIYI Token entry in the sidebar)

Include it in every asset-library request:

Authorization: Bearer sk-your-asset-library-KEY

Step 1: Upload the file and get a public URL

The asset file (an image; real-person assets also support video / audio) must first be reachable via a public URL. Pick either route:

A. You already have a public URL (your own CDN / file host) → skip to Step 2.

B. Upload to our storage (two calls: request a presigned upload URL → PUT the file):

# 1. Request a presigned upload URL
curl -X POST https://icover.ai/api/storage/presign \
  -H "Authorization: Bearer sk-your-asset-library-KEY" \
  -H "Content-Type: application/json" \
  -d '{"ext":"jpg","contentType":"image/jpeg"}'
# → { "code":0, "data": { "uploadUrl":"...", "publicUrl":"https://cdn.icover.ai/..." } }

# 2. PUT the file to uploadUrl (Content-Type must match the presign request)
curl -X PUT "the-uploadUrl-you-just-received" \
  -H "Content-Type: image/jpeg" \
  --data-binary @portrait.jpg
# Once it succeeds, publicUrl is your file's public address
# Same for video/audio: swap ext/contentType to mp4/video/mp4, mp3/audio/mpeg, etc.

Step 2: Ingest the asset

curl -X POST https://icover.ai/api/asset-library/assets \
  -H "Authorization: Bearer sk-your-asset-library-KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "imageUrl": "https://cdn.icover.ai/uploads/seedance/xxx.jpg",
    "label": "actor-A-front"
  }'
# → Raw Volcengine response: { ..., "Result": { "Id": "asset-20260702xxxx-xxxxx" } }

• groupId is optional: your default group is used / created automatically. To organize by character: first POST /api/asset-library/groups {"name":"actor-A"} to get a group ID, then include "groupId":"group-xxx" here.
• label is optional; it helps you identify assets in the web UI.

Step 3: Poll until Active

Ingest is asynchronous (about 13 seconds per image, no SLA). Poll with the returned Id:

curl https://icover.ai/api/asset-library/assets/asset-20260702xxxx-xxxxx \
  -H "Authorization: Bearer sk-your-asset-library-KEY"
# → Result.Status == "Active" means ready; "Failed" means re-upload

We suggest polling every 3 seconds; treat 90 seconds without Active as a timeout to investigate.

Step 4: Generate the video with the asset ID (via APIYI)

Write the asset ID as asset://<Id> and call APIYI with your own APIYI SeeDance2 token (not the asset library KEY):

curl -X POST https://api.apiyi.com/seedance/api/v3/contents/generations/tasks \
  -H "Authorization: Bearer sk-your-APIYI-token" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedance-2-0-260128",
    "content": [
      {"type":"text","text":"The character in Image 1 smiles at the camera, slow push-in, natural light"},
      {"type":"image_url","image_url":{"url":"asset://asset-20260702xxxx-xxxxx"},"role":"reference_image"}
    ],
    "ratio":"16:9","duration":5,"resolution":"720p"
  }'
# Returns a task id; poll GET .../tasks/{id} until status=succeeded, then read content.video_url

Refer to the asset as “Image 1” in the prompt — do not write the raw asset ID in the prompt.

For model selection, pricing, and resolution tables see the Seedance 2.0 Overview; for full generation parameters see the Video Generation API. For a complete runnable end-to-end script (upload → ingest → generate → download), see the Asset Reference Guide.

Full endpoint reference

Endpoint	Method	Description
/api/storage/presign	POST	Request a presigned file upload URL {ext?, contentType?}
/api/asset-library/groups	POST / GET	Create a group {name, description?} / list your groups (including real-person groups)
/api/asset-library/assets	POST / GET	Ingest {groupId?, imageUrl, label?, assetType?} (assetType optional: Image / Video / Audio, default Image) / list assets (?groupId=, ?pageNumber=, ?pageSize= all optional; defaults to page 1, 100 per page, pageSize capped at 100)
/api/asset-library/assets/{id}	GET / DELETE / PATCH	Check status / delete / update label {label}
/api/asset-library/real-person/sessions	POST / GET	Start a real-person verification {name?} → returns the H5 verification link and query credential / list your verification sessions
/api/asset-library/real-person/sessions/{id}	POST / PATCH / DELETE	Query the verification result (returns the real-person asset group GroupId on success) / rename {name} / delete the record
/api/asset-library/records	GET	Your assets + real-person verification archive (web UI data source)

Response conventions: asset-library endpoints pass through the raw Volcengine JSON on both success and failure (lists are filtered to your own assets); our own errors are plain text prefixed with [client] (400/401/403/404/502); records, the GET/PATCH/DELETE of real-person/sessions, and asset PATCH return {code, message, data} JSON (code 0 = success).

Real-face assets (fully automated API)

Real-person portraits require the person being filmed (the actor) to complete a one-time liveness verification, anchoring portrait-rights ownership at the source. The entire flow is now fully API-driven; the web UI’s “Real-Person Verification” tab is the same flow with an interface:

Step 1: Start a verification session and get the H5 link

curl -X POST https://icover.ai/api/asset-library/real-person/sessions \
  -H "Authorization: Bearer sk-your-asset-library-KEY" \
  -H "Content-Type: application/json" \
  -d '{"name":"actor-A"}'
# → Raw Volcengine response: { "Result": { "BytedToken":"2026...", "H5Link":"https://ark.volcengine.com/..." } }

• Send the H5Link to the actor to open on their phone (or turn it into a QR code to scan). They log in to their personal Volcengine account and complete the liveness verification. Lighting / camera angle can cause a failed attempt — just retry.
• BytedToken is the query credential; we store it with the session. GET /api/asset-library/real-person/sessions lists your sessions at any time (including id / status / h5Link).

Step 2: Query the result once the actor finishes

curl -X POST https://icover.ai/api/asset-library/real-person/sessions/{session-id} \
  -H "Authorization: Bearer sk-your-asset-library-KEY"
# Verified → { "Result": { "GroupId": "group-xxxx" } }  ← the actor's dedicated real-person asset group
# Not yet  → 404 NotFound.<token> (raw Volcengine text; this is normal — query again after verification)

Once you get the GroupId, the session status turns authorized and the real-person asset group is automatically archived to your account (it also shows up under the web UI’s asset groups). Note: a pending verification returns the same NotFound as an expired credential — the two cannot be distinguished. A link left unused for a long time may expire; just start a new session.

Step 3: Submit assets to the real-person group

Same ingest endpoint as virtual avatars — just pass the real-person group’s groupId. Images, video, and audio are supported:

curl -X POST https://icover.ai/api/asset-library/assets \
  -H "Authorization: Bearer sk-your-asset-library-KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "groupId": "group-xxxx",
    "imageUrl": "https://cdn.icover.ai/uploads/seedance/xxx.jpg",
    "label": "actor-A-front-full-body",
    "assetType": "Image"
  }'

Then poll until Active as usual and generate videos with asset://<Id> (Step 4 is unchanged).

Real-person asset rules and formats:

• One real-person group holds one person only; the same actor in different styling reuses the same group — no re-verification needed.
• Every upload goes through a face-consistency check (videos are sampled every second, and every sampled frame must pass). Side profiles, multiple people, or blurry material will fail — use clear frontal material.
• Images under 30MB; video mp4 / mov, 2–15 seconds, ≤50MB, aspect ratio 0.4–2.5; audio mp3 / wav, 2–15 seconds, ≤15MB.

Notes

Treat asset:// IDs as secrets: assets are isolated at this service layer (protected against listing and deletion), but Volcengine cannot authorize by ID — if an ID leaks, other callers on the same channel can reference it in generation requests. See the FAQ below for the full ownership and isolation model.

• Image URL lifetime: the preview URLs returned by query / list endpoints are temporary addresses valid for about 12 hours — do not cache them long-term; the asset ID itself is permanent.
• Rate limits (Volcengine account level): status query 100 QPS, ingest and other operations about 10 QPS. Control concurrency and retry on failure.
• Web UI status sync: after ingesting via API, if the status has never been queried, the web archive may show “processing”; open the asset list page and click “Refresh List” to sync the real status.

Volcengine official references (copy into your browser): private asset library guide volcengine.com/docs/82379/2333565, real-person asset onboarding volcengine.com/docs/82379/2315856.

FAQ

How is the asset library different from passing a reference image directly?

Direct reference images cannot contain photorealistic human faces (the anti-deepfake filter rejects them). Once a portrait is ingested as a trusted asset, its asset:// ID can be reused across any number of generation tasks, keeping the character’s face and clothing consistent across episodes and shots — ideal for serialized content such as animated drama, short drama, and IP characters. Real people can appear on camera too: just complete the real-person verification flow described above.

Does the asset library follow the account or the KEY? How is asset isolation implemented?

In one line: icover.ai isolates the assets; APIYI handles the unified calling — the generation side only checks the asset ID, and holding an ID is enough to reference it.

It follows the account. The underlying architecture: all icover.ai users share a single unified APIYI Volcengine account, and the asset library belongs to that account; icover.ai adds a layer of per-account isolation at the service level — each account can only list / query / delete its own assets and cannot see anyone else’s asset IDs.

KEYs provide no isolation: multiple Asset Library KEYs under the same account are equivalent and access the same library. If you need asset isolation (for example, separating data across customers or business lines), register a separate icover.ai account for each party and create KEYs under each — creating a new KEY under the same account does not isolate anything.

Mind the security boundary: this isolation covers listing / querying / deleting, but Volcengine cannot authorize by ID — once an asset:// ID leaks, other callers on the same channel can reference it in generation requests. Treat asset IDs as secrets.

Are assets stored on APIYI / icover.ai's own servers?

No. Whether through the web UI or the API, the asset library essentially forwards everything to Volcengine’s official asset library endpoints: image / video / audio content goes straight to Volcengine for processing, which returns only an asset:// ID. We don’t store or log the raw asset content ourselves — we only keep that ID and its account ownership (used for the per-account isolation described above).

Real-face assets carry an extra hard requirement: the person being photographed must complete live face verification in person (signing into their own Volcengine account to do face recognition), which locks down portrait-rights ownership at the source — no one else can complete this verification on their behalf. See “Real-face assets” above for the full flow.

How do multiple departments / teams within a company share or isolate the asset library?

Share one library (recommended, simplest): one account + one KEY is enough. Manage asset IDs centrally in your own system and distribute the asset:// IDs to each department — holding an ID is all it takes to reference it in a video-generation request (video generation uses each department’s own APIYI SeeDance2 token, unrelated to the Asset Library KEY). You can also create multiple KEYs under the same account and hand them to different departments (convenient for credential rotation / revocation) — these KEYs still access the same library.

Isolate between departments: register a separate icover.ai account for each department, each with its own KEYs. Note that isolation covers listing / querying / deleting — a leaked asset ID can still be referenced, so avoid spreading IDs casually even across departments.

Is this the full-capability SeeDance 2.0?

Yes. APIYI’s SeeDance2 channel runs the official full-capability doubao-seedance-2-0-260128 — model parameters, resolutions, and durations are identical to the official offering, with nothing cut down. See the Seedance 2.0 Overview for model details and pricing.

Does an AI-generated photorealistic portrait count as a real person? Does uploading imply authorization?

No, it does not count as a real person. An AI-generated photorealistic portrait with no real-world counterpart (for example, a character generated with a model like Nano Banana) is a virtual avatar — it goes straight through the virtual-avatar ingest flow with no authorization step. Only a photo of a real, existing person counts as a real face — for those, uploading does NOT imply authorization; the virtual-avatar ingest channel rejects them, and the person being filmed must complete liveness verification (see the real-face section above).

The three kinds of character material:

Material type	Examples	How to use
Anime / stylized characters	Anime, cartoon, 3D-cartoon characters	No photorealistic face, generally not blocked: use a public URL / Base64 reference image directly, no ingest needed
Virtual avatar (AI-generated photorealistic portrait)	Model-generated, no such person exists (see example below)	Use the virtual-avatar ingest on this page and reference the asset:// ID
Real face	Photos of celebrities, models, or the user themselves	Use the fully automated real-person verification API flow; usable once the person completes liveness verification

Virtual-avatar example: an AI-generated photorealistic portrait with no real-world counterpart. Put the frontal face close-up plus full-body front / side / back views into one asset group for the best character consistency

Do lifelike digital humans (virtual avatars) require review?

No. Virtual-avatar ingest is fully automatic — no manual review, no authorization step: after upload, the system preprocesses the image automatically, the status turns Active in about 13 seconds, and the asset:// ID is immediately usable for video generation. Only real-face assets require the person being filmed to complete liveness verification (see the real-face section above).

Can assets already ingested / verified with another provider be migrated over?

They cannot be reused directly. Both the Volcengine asset library and real-person verification are bound to the underlying account: an asset:// ID obtained through another provider belongs to that provider’s Volcengine account, and referencing it through the APIYI channel returns asset not found. Assets must be re-ingested with this service.

• Virtual-avatar assets: these can be migrated programmatically in bulk — write a script that re-uploads the original images through the API on this page, then update the old IDs in your system to the new asset:// IDs. We recommend keeping an asset-ID mapping layer in your own system: your business data stores only your internal IDs, so switching providers later only means updating the mapping.
• Real-person verified assets: real-person verification is also account-bound and cannot be migrated — re-verification is required: the person being filmed must complete liveness verification again (see the real-face section above).
• Advice for consumer-facing products: with a large stock of existing assets, migrate virtual assets silently via a backend job — users won’t notice; for the real-person verified portion, prompt users to re-verify on the occasion of a system upgrade or new version release, which feels much more natural.

Contact us

If you run into any issue during integration (ingest failures, real-person verification, batch integration, production token applications, etc.), feel free to reach out — contact details are on the api.apiyi.com homepage.