Skip to main content

Research Papers

Open-access research from Thirdrez Labs by Isaías Reis Verdin ORCID iD iconhttps://orcid.org/0009-0003-4031-7467 — founder of the Thirdrez ecosystem and author of the Kinetiq Engine and MotionPrint Protocol. Both papers are published on Zenodo with permanent DOIs and are free to read, cite, and share.

Kinetiq Engine — Hybrid Generative Motion Synthesis​

A technical overview of the Kinetiq architecture: solving the "Gelatin Problem" via Ecological Interaction Data and a client-side WebAssembly refinement pipeline.

Author: Isaías Reis Verdin (Thirdrez Labs) Published: December 2025 DOI: 10.5281/zenodo.17849581 Type: Technical Whitepaper

Summary​

Generative motion models can produce plausible movement from text prompts, but their output rarely survives a real production pipeline. Foot sliding, joint hyperextension, mesh interpenetration, and root drift accumulate across long sequences — artifacts the paper groups under the "Gelatin Problem": motion that lacks weight and deterministic contact.

Kinetiq Engine takes a hybrid approach: a generative backbone for variation, paired with a deterministic refinement layer that runs client-side in WebAssembly. The browser becomes the cleanup studio — no installs, no uploads, no desktop GPU required. The whitepaper covers:

  • A formal characterization of the Gelatin Problem and its biomechanical roots
  • The Ecological Interaction Data training paradigm — how naturalistic, intent-driven motion differs from controlled lab mocap
  • The four-stage Kinetiq pipeline (synthesis → physics refinement → style adaptation → export)
  • Real-time, in-browser refinement at 60 FPS on consumer hardware
  • A privacy-preserving architecture where user animations never leave the device during cleanup

📄 Read the Whitepaper →

Citation​

@misc{verdin2025kinetiq,
  title  = {Kinetiq: Hybrid Generative Motion Synthesis via Ecological Interaction Data},
  author = {Verdin, Isaías Reis},
  year   = {2025},
  month  = {December},
  publisher = {Zenodo},
  doi    = {10.5281/zenodo.17849581},
  url    = {https://doi.org/10.5281/zenodo.17849581}
}

MotionPrint Protocol — Cryptographic Provenance for 3D Motion Data​

A privacy-first protocol for authenticity in generative 3D pipelines: client-side Ed25519 signatures, embedded directly in the file. Zero servers, zero uploads, mathematical trust.

Author: Isaías Reis Verdin (Thirdrez Labs) Published: December 2025 DOI: 10.5281/zenodo.17843580 Type: Protocol Specification

Summary​

As generative AI accelerates 3D content production, attribution and chain-of-custody are increasingly hard to verify. Existing 3D watermarking techniques (spectral embedding, vertex displacement, texture watermarks) are destructive, format-specific, and easily stripped by routine workflow operations. Worse: most verification systems force creators to upload their work to a third-party server.

MotionPrint takes the opposite approach. It defines a non-destructive provenance schema that signs motion files using Ed25519 elliptic-curve signatures, embedded in spec-compliant metadata containers (extras for glTF/GLB, header comments for BVH). Verification runs entirely in the browser via WebAssembly — the file is never transmitted. The paper covers:

  • A canonical, deterministic JSON payload schema for signed provenance
  • Format-specific, non-destructive embedding strategies (glTF/GLB and BVH)
  • A client-side verification pipeline using the audited @noble/ed25519 library
  • Sub-10 ms verification latency with under 400 bytes of overhead
  • Automated PDF Certificate of Authenticity generation with verification QR

The framing matters: MotionPrint is provenance (attestation), not DRM (restriction). It tells you who signed a clip and proves the bytes are unchanged — without locking anyone out, calling home, or tracking creators.

📄 Read the Protocol →

Citation​

@misc{verdin2025motionprint,
  title  = {MotionPrint: A Privacy-Preserving Cryptographic Provenance Protocol
            for 3D Motion Data Using Ed25519 Digital Signatures},
  author = {Verdin, Isaías Reis},
  year   = {2025},
  month  = {December},
  publisher = {Zenodo},
  doi    = {10.5281/zenodo.17843580},
  url    = {https://doi.org/10.5281/zenodo.17843580}
}

Collaborate​

We welcome conversations with academic institutions, independent researchers, and labs working on:

  • Motion quality metrics and perceptual studies
  • Cryptographic provenance for generative content
  • Real-time, on-device inference and refinement
  • Cross-skeleton motion transfer

Contact: [email protected]

All Thirdrez research is released under open-access licenses on Zenodo and conducted in line with ethical AI principles and data-privacy regulations.