4dgs-human-avatar-comfyui-nodes-dataset

Preparing to release the ultimate 4DGS module for ComfyUI. Take your seats.

License ComfyUI Status

Faithful volumetric video of real performances — surround coverage from multi-camera capture, engineered for measurable stability.

---

---

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.

**→ [Go to the landing page and add yourself to