arias-deepfake-investigation — anchor impersonation deepfake
Regional news anchor impersonation bundle: PRNU-matched authentic stills, face-swap composite, copy-move scene, GAN-grid synthetic headshot, natural vs spliced voice WAV, and six-frame temporal metrics strip. Fully synthetic.
what this proves
- all eight deepfake-investigation primary engines produce deterministic, fixture-locked output — verified by
npm run check:flagship(280/280 fleet · 8 for this scenario). - every output is generated 100% locally in your browser — preserve originals, never re-export through chat.
- video temporal metrics, face-swap boundaries, GAN-grid fingerprints, synthetic voice artifacts, audio splice seams, ELA tamper regions, PRNU sensor correlation, and copy-move scene forgery surface without uploading evidence.
primary engines locked to this fixture
build the case binder
runs all eight primary engines on the synthetic evidence zip and opens a self-contained html binder. uses the default binder renderer for deepfake investigation — no upload.
runs all 8 primary engines locally on the synthetic evidence zip · opens a self-contained html binder · no upload
download the synthetic evidence
MIT-licensed, fully synthetic. includes PRNU-matched authentic stills, face-swap composite, copy-move scene with ELA tamper region, GAN-grid synthetic headshot, natural vs spliced voice WAV, six-frame temporal metrics strip, and incident overview.
built deterministically from scripts/fixtures/build-arias-deepfake-investigation.mjs. seed: arias-deepfake-investigation:v1.
methodology
deepfake impersonation is multimodal — preserve originals before any platform re-export. walk video temporal → face swap → GAN fingerprint → synthetic voice → audio splice → ELA stills → PRNU sensor match → copy-move scene integrity. read the full deepfake investigation (video / audio / image) guide →
after the playbook
run each primary locally — or export findings from the binder — then drop every csv/json into fatcousin-multi-tool-super-timeline-correlator. one timestamp-sorted timeline across video temporal drift, face-swap boundaries, synthetic voice seams, and PRNU sensor mismatch — still zero upload.