The fatal flaw of AI video generation is hallucination. When models guess architectural blueprints, historical lore, or scientific telemetry, the output is fiction.
Introducing Chronicle DKG: the first verifiable, knowledge-grounded generative media studio.
We connected OriginTrail's Decentralized Knowledge Graph directly to Livepeer's Agent Creative MCP.
The workflow:
1. Inbound SPARQL Knowledge Grounding
Before generating a single frame, Chronicle executes SPARQL queries against OriginTrail DKG v8 to extract verified knowledge triples (telemetry, architectural blueprints, canon records).
2. Fact-Constrained Generation on Livepeer
Extracted triples strictly constrain diffusion prompts dispatched to
@Livepeer's Agent MCP. Decentralized GPU nodes render visual evidence bounded by verified truth, eliminating hallucinations.
3. Graph Inspector & Live Merkle Verification
Explore the interactive Graph Inspector at 60 FPS. Scrub real-time SPARQL assertions, inspect entity claims, and trigger targeted shot re-rolls while preserving narrative ground truth.
4. Permanent UAL Minting on NeuroWeb
Finalized shots are minted as immutable Knowledge Assets on NeuroWeb OTP:2043, binding C2PA manifests and orchestrator node signatures to a permanent UAL.
Zero hallucination. Cryptographic provenance.
Built for the
@Livepeer x
@origin_trail Track in
@Embody_Net Livepeer Agent Hackathon.
GitHub:
github.com/Redd1xx/chronicle…
Live App:
chronicledkg.lol/