Documentation/Practical guides

Practical guides

Open-source AI watermarking: libraries, a kit or a managed API?

For an AI image or video app, first choose how much of the labeling workflow your team wants to assemble and operate. A watermark engine, C2PA library, connected self-hosted kit and managed service solve different parts of the job.

Compare the operating models

Approach You get Your team still handles
CAI / C2PA libraries Building blocks for signed Content Credentials Watermark/recovery choices, record semantics, final-rendition workflow and product operations
Adobe TrustMark / Meta VideoSeal Watermarking methods and research implementations Integration, signing, verifier/registry, key management, qualification and legal mapping
Ariadne Mark A connected CPU image/video SDK, verifier, registry and EU/California implementation guides Production identity/TSA, hosting, delivery tests and your product-specific actions
Managed provenance or labeling API The provider's hosted integration and service scope Vendor selection, contract, data flows, operating costs, portability and applicability

This is a comparison of operating models, not a robustness ranking or an assertion that the products are legally equivalent. Consult the original CAI tools, TrustMark, VideoSeal and the described Trufo platform for their current scope.

What does the ready kit save you from assembling?

For a self-hosted AI image/video labeling project, the integration work extends beyond embedding a watermark. Ariadne connects final-export marking, truthful C2PA records, read-back checks, verification results, signed reports and ID-to-manifest recovery. It also supplies source-linked EU/California requirements and reproducible CPU measurements.

You still own infrastructure, production credentials and tests on your delivery pipeline. A managed API can be the better fit when vendor-operated infrastructure matters more than local processing. Direct libraries can be better when you need a different watermark or record architecture.

Your priority Inspect first
Free open-source image/video watermarking on a CPU backend The connected first integration
Full control of your watermark or provenance architecture The component libraries and their own licences above
Hosted operation and a vendor service contract A managed provider's actual scope, data flow and quote
Comparing the complete cost Compute, signing and operations worksheet

When is a self-hosted kit useful?

Consider Ariadne when your team wants to keep media processing in its own backend and start with connected tools instead of assembling every component. The qualified workflow runs on Linux x86-64 CPU and supports the documented image/SDR MP4 formats. The quickstart gets you to a first export; samples and measurements make that result inspectable.

Plan your production signing, verifier hosting and registry persistence. A project-code licence without a per-export fee does not eliminate your operating costs.

For this release, current image and video watermarking use Adobe TrustMark Q; Meta VideoSeal is a legacy option. This differs from composing a new VideoSeal-based pipeline yourself. The 0.7.3a4 CPU measurements describe Ariadne's actual export path, not a claim about the speed of VideoSeal or a private commercial engine. Review all twelve pre-integration questions before comparing scopes.

When should I compose the components myself?

Direct libraries can fit a team that needs tighter control of record schemas, signing integration, media handling or a specialized watermarking method. Evaluate those components against the exact transformations and verification requirements of your product. Include the work needed to connect them, operate keys and maintain an accessible detector.

When should I evaluate a managed service?

A hosted provider can fit a team that wants vendor-operated signing or infrastructure and is comfortable with its data flow, terms and cost model. Ask what stays local, which bytes are uploaded, how verification works and what happens to records if you leave. We have not audited a managed provider's private engine, frame cadence, service reliability or unit economics.

Questions to ask before choosing

  1. Can I download and independently verify an export from the exact version offered?
  2. Which file formats, dimensions, durations and transformations were tested?
  3. What is included: watermarking, C2PA signing, timestamps, verifier, registry or only one layer?
  4. What are the cold/warm CPU, memory and operating costs for my workload?
  5. Who controls the keys and recovery records, and can I retain the evidence?
  6. Which product facts, disclosures and operational duties remain mine?

Use the requirements matrix to evaluate coverage, licence inventory for commercial-use terms, and verification results to interpret what an output actually establishes.