# DreamTechAI/Direct3D-S2

[NeurIPS 2025] Direct3D‑S2: Gigascale 3D Generation Made Easy with Spatial Sparse Attention

Repository: https://github.com/DreamTechAI/Direct3D-S2
Canonical: https://ross.abutalabs.com/products/direct3d-s2
Homepage: https://neural4d.com/research/direct3d-s2
Language: Python
License: MIT
License Family: permissive
Topics: 3d-aigc, 3d-generation, 3d-models, 3d-reconstruction, 3d-representation, image-to-3d, genai-3d
Last push: 2025-09-26T06:22:00+00:00

## Health v2 (maintenance only)
Score: 29/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 44, release rhythm 8, longevity 33
- inputs: {"age_days": 467, "days_push": 341, "days_rel": 460, "gap_med": null, "n_releases_24m": 1}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1274, forks 107 (observed 2026-08-28T04:04:12.816512+00:00)

## What it is
Direct3D-S2 is a research framework for high-resolution 3D shape generation from images, built on sparse volumetric representations and a novel Spatial Sparse Attention mechanism for Diffusion Transformers. It includes a unified sparse-volume VAE and pretrained models enabling gigascale (1024^3) 3D generation on as few as 8 GPUs.

## Use cases
- generate a 3D model from a single image
- create high-resolution 3D meshes for games
- convert photos into 3D assets
- research sparse attention for 3D diffusion models
- train 3D generative models on limited GPUs
- reconstruct 3D shapes with SDF volumetric representations

## When to choose
- you need image-to-3D generation with very high volumetric resolution
- you are researching efficient attention for sparse 3D data or DiT models
- you want a state-of-the-art open 3D generation model with released weights
- you have GPU resources and want to run or fine-tune 3D AIGC models

## When to avoid
- you need a polished end-user 3D creation app rather than research code
- you have no GPU available, since inference requires significant VRAM
- you need text-to-3D or full texturing pipelines out of the box
- you need production support or a stable API contract

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, image-processing, graphics
- domain: artificial-intelligence, machine-learning, graphics
- platform: python
- tags: 3d-generation, image-to-3d, sparse-attention, diffusion-transformer, sdf, 3d-reconstruction, research-model, neurips-2025, game-development, gpu, linux

## Member repositories
- DreamTechAI/Direct3D-S2 (main) score 29

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:12.816512+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T05:03:04.685866+00:00, confidence not recorded.
  - readme: https://github.com/DreamTechAI/Direct3D-S2 (fetched 2026-08-28T04:04:12.816512+00:00, sha d52103260559)
  - homepage: https://neural4d.com/research/direct3d-s2 (fetched 2026-08-29T12:14:25.985440+00:00, sha 6a2cba36ae16)
  - site_page: https://www.neural4d.com/pricing (fetched 2026-08-29T12:14:25.995207+00:00, sha 41dce04ae9c5)
- Data as of 2026-08-30T08:39:29.467469+00:00.