4dgs-human-avatar-comfyui-nodes-dataset
Faithful volumetric video of real performances — surround coverage from multi-camera capture, engineered for measurable stability.
Preparing to release the ultimate 4DGS module for ComfyUI. Take your seats.
Faithful volumetric video of real performances — surround coverage from multi-camera capture, engineered for measurable stability.
<p align="center"> <a href="https://comfyui-4dgs-volumetric-node.eu/#waitlist"> <img src="https://comfyui-4dgs-volumetric-node.eu/video/hero-poster.jpg" alt="4DGS Volumetric — Shramko 4D" width="720" /> </a> </p> <p align="center"> <strong> <a href="https://comfyui-4dgs-volumetric-node.eu/">Watch the demo & join the waitlist →</a> </strong> </p>
There’s nothing to run here yet — and that’s why we need you
This repo will host ComfyUI nodes (Shramko-Volumetric-Bridge) and human scan datasets for 4D Gaussian Splatting. The module and data are being prepared. There is no downloadable build or dataset yet. If you want to be first in line when we ship, add yourself to the closed waitlist on our landing page.
We’re all tired of noisy, heavy 4DGS and non-commercial licenses
Most of today’s 4D Gaussian Splatting stack is research code under non-commercial licenses, with no stable end-to-end path from a real captured performance to a playable volumetric clip. Feed-forward single-image tools (e.g. Apple SHARP — a genuinely impressive static 3DGS generator whose released weights are research-only) solve a different problem: they generate a scene from one photo, they don’t faithfully replay a real performance. We’re building the missing piece: faithful 4DGS of a real captured performance, delivered through ComfyUI, on a commercially clean stack.
Our target characteristics (targets, not shipped results — derivations and verification status for every number live in Research-docs/v7):
| Target | Honest fine print | |--------|-------------------| | Faithful surround view from multi-camera capture | single-camera captures are front-faithful; nobody can observe an unseen back — learned back-fill is a separate, clearly-labeled research track | | Measurable temporal stability | acceptance metrics (held-out-camera PSNR/LPIPS, flicker checks) are defined in Research-docs/v7 — no “absolute” claims | | Commercially clean pipeline | built on Apache-2.0 components (gsplat, Wan 2.1) with a strict license wall; details in LEGAL-LICENSING | | Compact temporal files | published research codecs already reach ≈0.35–0.7 MB/frame (DualGS, NVIDIA QUEEN) — our target is that class of size on a commercially clean codepath |
Why the waitlist?
Building this module — and capturing, processing and legally clearing large volumes of volumetric data — is a huge investment (the honest current status of data and code is tracked in DATASET-STATUS). Before the final push, we need to see real interest. Your signup is a signal that we’re building what the market actually needs. We also need to know: will you run nodes locally, or do you need a cloud API? Leave your details on the landing page — when we’re ready, you’ll get access first.
Reserve early access. Launch is around the corner.
Your waitlist signup is a vote for this product and a ticket to the front row. Everyone who joins at this stage gets exclusive terms right after launch — enough credits for plenty of tests.
<p align="center"> <a href="https://comfyui-4dgs-volumetric-node.eu/#waitlist"> <img src="https://img.shields.io/badge/Join_the_closed_Waitlist-00D4FF?style=for-the-badge&labelColor=0A0A0A" alt="Join the closed Waitlist" /> </a> </p>→ Go to the landing page and add yourself to the waitlist
What will be in this repo (when it’s ready)
- ComfyUI nodes — Shramko-Volumetric-Bridge: Ingest, Processing, Visualization, Export. 4DGS creation as intuitive as regular video generation.
- Free dataset — Human scan data for non-commercial use and research (CC BY-NC-4.0), hosted on Hugging Face.
- Enterprise dataset — planned: large-scale, consent-documented human capture for training your own AI models under a commercial license (production gates and scale are defined in Research-docs/v7/VARIANT-3; current status: DATASET-STATUS).
Until then, star this repo and join the waitlist so we can notify you at launch.
Development & Research Documentation
Research and implementation docs (architecture, loss functions, kinematics, deployment, legal strategy, etc.) are in the Research-docs folder. Start from Research-docs/v7 — the audited, three-variant plan where every claim carries a verification tag (see its CLAIMS_REGISTRY). v5/v6 remain for provenance and implementation detail. This documentation was produced with AI assistance; v7 claims are individually verified and tagged, earlier versions may contain inaccuracies.
Links
- Landing page (demo, video, waitlist): comfyui-4dgs-volumetric-node.eu
- Waitlist form: comfyui-4dgs-volumetric-node.eu/#waitlist
Andrii Shramko · Engineer · Inventor · 4DGS Researcher · LinkedIn · Book a video call
Repository structure (for later)
├── Research-docs/ # Development & research documentation (v5, v6, v7 — start at v7)
├── comfyui_nodes/ # Shramko-Volumetric-Bridge nodes (when released)
├── models/ # Model weights & instructions
├── dataset/ # Free & enterprise dataset docs and scripts
├── docs/
└── examples/
Licenses: Code — Apache 2.0. Free dataset — CC BY-NC-4.0. Enterprise — see terms.
Development and research documentation in Research-docs was produced with AI assistance and is shared for reference only; it may contain inaccuracies or be outdated.