01 / SDK
Label at export.
Invisible watermarks, signed C2PA provenance and read-back checks for your image and video pipeline.
Integrate the SDK ↗Developer alpha · 0.7.2a1
Be creative.
Stay compliant.
Compliance · Provenance · Misuse Prevention
A DIY kit for EU and California
AI labeling compliance.
Watermarking, provenance and verification — ready to integrate.
↳ Invisible watermark
Built for teams shipping generative products.
Images & videoInvisible watermarksC2PA provenanceVerification & recovery
Start on the qualified runtime.
Linux x86-64 · Python 3.12 · CPU. Recommended: the Docker release bundle. PNG/JPEG images and supported SDR MP4 video.
Free code. Your operating costs.
Budget for compute, hosting and production signing or timestamps. See what to provision.
Check your role and market.
EU provider marking duties generally apply from 2 August 2026, with a transition for certain existing systems. Check EU and California scope and dates.
01 / FROM CREATION TO VERIFICATION
Keep the creativity.
Carry the origin.
An invisible watermark in the media.
A signed record alongside it.
A way to read both.
A VISUAL WALKTHROUGH
Embed a recoverable ID into your final image or video, then sign its C2PA provenance.
Read the integration guide ↗↳
An origin you can read.
The visible image remains the creative output. The labeling layer travels with the file.
This is an interactive illustration of the workflow, not a live media scan. Actual results distinguish recovered IDs, signatures, issuer trust and file binding.
02 / WHY ARIADNE
We did the groundwork.
You get the toolkit.
Requirements to research. Tools to connect. Behavior to test. We brought that work into one open-source labeling kit for images and video.
Explore the connected kit ↗010203
Requirements mapped.
EU and California duties connected to technical measures, sources and remaining operator actions.
Tools connected.
Watermarking, C2PA provenance, a verifier and a recovery registry — in one self-hosted workflow.
Behavior tested.
Versioned CPU tests and transformation experiments, with methods, results and known limits.
03 / ONE OPEN TOOLKIT
Three pieces.
One connected workflow.
Bring the tools into your product.
Run the workflow on your infrastructure.
02 / VERIFIER
Make origin readable.
A self-hosted upload UI and API for checking supported marks, provenance records and signed reports.
Explore verification ↗03 / REGISTRY
Reconnect the history.
Link recoverable IDs to stored provenance records when supported watermarks survive metadata loss.
See recovery & storage ↗04 / VIDEO, AT YOUR PACE
Choose the work
your pipeline needs.
Economy modes reduce watermark inference.
Resilient computes a fresh mark for every frame.
LOWER WATERMARK WORK
A little inference.
A full video export.
Fresh watermark computation in 125 ms windows every five seconds. Most frames bypass inference; the full video still passes through encoding.
See the protocol & trade-offs ↗FRESH WATERMARK WORK10-SECOND TIMELINE
0s5s10s
"video_profile": "economy_5s"Short excerpts can miss the marked windows. Check the preset against your content and delivery transformations.
05 / OPEN CODE. VISIBLE EVIDENCE.
Don’t take a
black box for it.
Read the code. Inspect the test protocols.
Bring your own content and improve the toolkit.
Project codeApache-2.0
DeploymentYour infrastructure
Media processingCPU supported
Current release0.7.2a1 / developer alpha
Third-party engines keep their own licenses ↗A FEW PRACTICAL ANSWERS
Good questions.
Clear answers.
Is it free to use?+
The project code is Apache-2.0, with no Ariadne per-export fee. You pay for compute, hosting and any production signing or timestamp services. Third-party components and model weights keep their own license terms. Plan the operating costs ↗
Does the SDK make my whole product compliant?+
It supplies marking, provenance, verification and recovery tools. Use the legal guides to check applicability, configure your real provider identity, deploy the verifier and address any required disclosure or operating obligations.
What does misuse prevention mean here?+
Recoverable IDs and provenance records support attribution and investigation of suspicious reuse. They do not moderate generation, prevent every misuse or establish that a scene is true. Understand the verification results ↗
Do I need a GPU?+
The qualified workflow runs on CPU. Video still needs decoding and encoding, and the profile affects inference work. Use the versioned benchmarks and test your own content before sizing production.
Is the verifier a hosted Ariadne service?+
The kit includes software you can self-host: an upload UI, API and registry. The image playground on this site runs real Mark and Verify operations. Your production verifier is a separate deployment.
BUILT FOR THE NEXT THING YOU MAKE.
Create something
worth tracing back.
Build AI labeling into your next release.
Start with the SDK ↗