ModelArk Artifact Verification

ModelArk Artifact Verification: Provenance Checks for Seedance and Seedream

What is ModelArk Artifact Verification?

ModelArk Artifact Verification is a provenance API that checks whether an image or video was generated by a BytePlus ModelArk model — the Seedance video family and Seedream image family from ByteDance. You pass a single public media URL, and the service returns a structured verdict: whether the file carries technical features associated with ModelArk artifacts and, when possible, the specific model it attributes the file to (for example, dreamina-seedance-2-5). It wraps BytePlus's official artifact-verification service behind one synchronous call. Every result is automated analysis and is for reference only.

Key Features

• •Single input: a public image or video URL — no prompt engineering required.
• •Structured JSON output: is_official (true, false, or null), a stable verdict label (official, not_official, or undetermined), a detected model_name, resource_type, and a query_id for support.
• •Broad media support: images in png, jpg, jpeg, heic, webp up to 20MB, and video in mp4, mov, avi up to 100MB.
• •Synchronous response, typically in about 10 to 13 seconds, with a human-readable message explaining each verdict.

Best Use Cases

• •Content moderation and trust-and-safety pipelines that need to flag ModelArk-generated media.
• •Newsrooms, marketplaces, and platforms confirming whether an asset came from Seedance or Seedream.
• •Compliance and audit trails that log each provenance check with a query_id.
• •Creators confirming their own ModelArk outputs are correctly attributed.

Interpreting Inputs and Output Quality

Use direct object-store links that end in a file extension for the most reliable uploads. Treat only verdict == official as a positive match. An undetermined result (is_official is null) is the normal answer for ordinary photos and non-ModelArk media, because no identifiable ModelArk traits were found. A non-match never rules out a ModelArk source, and this tool does not detect other vendors' models or act as a general AI-image detector.

How to Use ModelArk Artifact Verification

ModelArk Artifact Verification answers one question: was this image or video generated by a BytePlus ModelArk model (Seedance or Seedream)? You send a single public media URL and read back a structured verdict, so it slots cleanly into moderation, provenance, and audit workflows.

How it works

Pass a public URL in content_url. The service analyzes the file and returns is_official (true, false, or null), a stable verdict label (official, not_official, or undetermined), a detected model_name when available, resource_type, a plain-language message, and a query_id you can quote to support. There are no other parameters to tune — reliability comes from the input you provide, not from settings.

Recommended Settings

• •content_url: use a direct object-store link that ends in a file extension (for example, a .mp4 or .jpg URL). Images must be png, jpg, jpeg, heic, or webp up to 20MB; video must be mp4, mov, or avi up to 100MB. Base64 and data URIs are accepted and uploaded for you, and redirecting links are followed server-side.
• •Reading results: treat only verdict == official as a positive match. Log the query_id for every check so you can trace a determination later.

Example files to verify

Each example below was run end to end at the exact input shown, and the media URL is what carried over from request to verdict.

• •ModelArk-generated video. content_url: "https://segmind-resources.s3.amazonaws.com/output/seedance-2.5-example-output.mp4". The output returned is_official: true with verdict: official and attributed the clip to dreamina-seedance-2-5 in about 12 seconds. Tip: expect a model_name on genuine ModelArk media, and store it alongside the query_id for your audit record.
• •Ordinary photograph. content_url: "https://segmind-resources.s3.amazonaws.com/input/modelark-artifact-verification-real-photo.jpg". A real camera photo returned undetermined with is_official: null, because no identifiable ModelArk traits were found. Tip: expect ordinary media to land on undetermined rather than not_official, and do not read that as proof the file is not AI-generated.

Further checks the model is built for: "verify a Seedream image export," "confirm a downloaded Seedance clip before publishing," or "batch-check a folder of user uploads for ModelArk provenance." These are suggestions to try, not runs we performed.

Remember that verdicts are automated and for reference only, and the service checks for ModelArk provenance alone — it is not a general AI-content detector.

Does this detect any AI-generated image or video?

No. It only checks for BytePlus ModelArk (Seedance, Seedream) provenance, not other generators.

What does undetermined mean?

The service found no identifiable ModelArk traits — the usual result for real photos and non-ModelArk media.

Which formats and sizes are supported?

Images up to 20MB (png, jpg, jpeg, heic, webp) and video up to 100MB (mp4, mov, avi).

Can it tell me which model made the file?

When it detects ModelArk artifacts it often returns a model_name, such as dreamina-seedance-2-5.

Is the verdict definitive?

No. Results are for reference only; a non-match does not prove a file is not from ModelArk.