markproof

Python client for the Markproof API — EU AI Act Article 50 watermarking and verification in one call. Thin wrapper over requests.

Is your AI output Article 50-ready? Drop a file into the free checker (no signup) to see whether it carries a watermark and signed Content Credentials.

Install

pip install markproof

Usage

from markproof import Markproof

client = Markproof(api_key="lt_live_...")

marked = client.mark("generated.png", generator_name="my-app", model_name="my-model-v1")
print(marked.watermark_code, marked.verify_url)
with open("generated-marked.png", "wb") as f:
    f.write(marked.file)

result = client.verify("generated-marked.png")
print(result.summary)

API

• Markproof(api_key, base_url=DEFAULT_BASE_URL) — base_url defaults to the production API; override for local dev.
• client.mark(file, filename=None, generator_name=None, model_name=None, response_format="file", visible_label=None, label_position=None) → MarkResult — file is a path (str/os.PathLike) or raw bytes (in which case filename is required). response_format="file" (default) returns the marked bytes inline (result.file/result.content_type); "url" returns a short-lived signed download URL instead (result.file_url, ~1 hour — download promptly rather than caching it). Either way you also get asset_id, watermark_code, verify_url, c2pa_manifest_label. visible_label (images only): True draws a visible "AI-generated" label (EU AI Act Art. 50(4)) before marking, or pass your own text (max 40 characters); label_position picks the corner ("bottom-right" default, "bottom-left", "top-right", "top-left").
• client.verify(file, filename=None) or client.verify(url=...) → VerifyResult — watermark/c2pa detection results, matched_asset if the watermark matches a record in Markproof's database, and a human-readable summary. Direct uploads are limited to about 4.5 MB; pass a public url (up to 25 MB) for bigger files, such as marked audio, which comes back as WAV (e.g. a batch item's file_url).
• client.get_asset(asset_id) → Asset — requires an API key on the same project that created the asset.

All methods raise MarkproofError (with .status and .body) on a non-2xx response.

Batches and webhooks

For volume, queue many files at once instead of calling mark per file. Each item counts as one mark.

created = client.create_batch(
    urls=["https://cdn.example.com/1.png", "https://cdn.example.com/2.png"],  # and/or files=["a.png", (data, "b.wav")]
    generator_name="your-app",
    webhook_url="https://your-app.example.com/hooks/markproof",  # optional
)
batch = client.wait_for_batch(created.batch_id)  # or get_batch(...), or wait for the webhook
for item in batch.items:
    print(item.index, item.status, item.file_url or item.error)

• client.create_batch(files=None, urls=None, generator_name=None, model_name=None, webhook_url=None) → CreatedBatch — files go up in one multipart request (keep the total under ~4.5 MB; use urls for more). Max items per batch: 10 Free, 100 Indie, 1,000 Pro.
• client.get_batch(batch_id) → Batch — status (queued/processing/completed/partial/failed), counts, and items with asset_id, watermark_code, verify_url, file_url (signed, ~1 hour; call again for a fresh one) or error.
• client.wait_for_batch(batch_id, interval=3.0, timeout=900, on_progress=None) — polls until the batch is final.

Verify webhooks with the raw request body and your signing secret from the dashboard:

from markproof import verify_webhook_signature

ok = verify_webhook_signature(
    request.get_data(),  # the exact bytes received, not re-serialized JSON
    request.headers.get("Markproof-Signature"),
    os.environ["MARKPROOF_WEBHOOK_SECRET"],
)

parse_webhook_event(...) does the same and returns the batch.completed event, raising WebhookSignatureError if the signature doesn't match. Event ids are stable across retries, so use them to drop duplicates.