Documentation/Install and integrate

Install and integrate

Install and produce your first marked export

SDK 0.7.3a4 developer alpha · documentation updated 2026-11-14.

Versioned benchmark and legal-source review dates below remain historical evidence; they are not a new qualification or legal approval of this candidate.

Use this path for your own generated JPEG/PNG or final SDR MP4 rendition. The supported media runtime is Linux x86-64/Python 3.12 on CPU. The record/profile core also supports Python 3.11–3.14. Synthetic audio and text marking are outside this visual kit; AAC may be preserved as original audio.

For one tested route from these samples to your own files, use the image + silent-video integration walkthrough.

Get one complete result first

Download Release 0.7.3a4, starting with START_HERE.txt. FIRST_EXPORT.sh verifies the SDK/sample artifacts and runs the image plus three-video-mode walkthrough. Inspect it before running. The default Docker build needs about 10 GB free; the wheel path uses less disk when its system prerequisites are already installed.

The release archive's bundled README describes its build-time state. For current installation and publication status, follow release START_HERE.txt and this page; the archived text is preserved with its original hashes.

The immutable runtime/source is 49578688710a8bae2373aaf9d47ff9521f566d4c; subsequent documentation changes do not replace those tested package bytes. Qualification and migration notes describe the exact scope.

Choose an installation

The source README contains the Docker build commands for the matching qualified source archive. The wheel guide provides an installation outside the source checkout:

python3.12 -m venv .venv
. .venv/bin/activate
python -m pip install --no-deps /path/to/ariadne_mark-0.7.3a4-py3-none-any.whl
ariadne setup media --fetch-models
ariadne runtime check

FFmpeg/ffprobe and Node.js/npm are prerequisites for that wheel path. Use ariadne setup video --fetch-models for video-only processing; it excludes Torch, NumPy, Python TrustMark and legacy decoders. Explicit setup may download files; processing does not install missing dependencies. A production TSA request is a separate, explicitly configured signing-time network operation.

Evaluate the complete local kit

In the provisioned environment:

ariadne kit init deployments/demo --development --publisher "Example Studio"   --system "Example Generator" --system-version "1.0"
ariadne kit check deployments/demo --runtime

Put a supported final file under deployments/demo/inputs/. Copy the generated batch.example.json to batch.json. Supply its real source/destination and actual generation time with timezone; choose a new output name. For video, choose economy_5s, economy_2s or resilient, and declare audio correctly. The generated kit README has the Docker/Compose commands and pinned edge setup. Alternatively, with this environment and kit on the host:

ariadne kit export deployments/demo deployments/demo/batch.json

A successful job produces the signed file, its .receipt.json and a registry entry. Publish the returned file, retaining the receipt internally. The original input is not a fallback publication if marking fails. A failed batch may have committed earlier successful jobs; inspect per-file receipts before retrying.

Check the result

ariadne verify deployments/demo/outputs/YOUR_OUTPUT.png --verification-scope current --registry deployments/demo/registry/assets.sqlite3

Development files intentionally show a development-signature headline. They do not become production credentials by trusting their test root. For the complete HTTPS UI/API run the generated Compose kit as described in the deployment guide. Pin its local TLS certificate; do not disable certificate verification.

Move to production

Follow production onboarding. One kit represents one publisher/system/version pair. Prepare crop, size and overlays before marking each rendition. Preserve the published bytes or qualify later transformations. For a generator already returning C2PA credentials, first read the upstream-input boundary.