# nv-tlabs/Difix3D

[CVPR 2025 Oral & Best Paper Finalist] Difix3D+: Improving 3D Reconstructions with Single-Step Diffusion Models

Repository: https://github.com/nv-tlabs/Difix3D
Canonical: https://ross.abutalabs.com/products/difix3d
Homepage: https://research.nvidia.com/labs/toronto-ai/difix3d/
Language: Python
License: NOASSERTION
License Family: other
Topics: diffusion-models, gaussian-splatting, nerf, 3d
Last push: 2025-06-28T01:31:31+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 28, release rhythm 35, longevity 35
- inputs: {"age_days": 495, "days_push": 432, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1266, forks 107 (observed 2026-08-28T04:04:11.188045+00:00)

## What it is
Difix3D+ is a research codebase from NVIDIA implementing a single-step diffusion model pipeline that removes artifacts from NeRF and 3D Gaussian Splatting reconstructions. It provides a diffusers-compatible inference pipeline plus training scripts for the Difix image enhancement model.

## Use cases
- remove artifacts from NeRF renders
- improve 3D Gaussian Splatting reconstruction quality
- enhance novel view synthesis with diffusion models
- clean up underconstrained regions in 3D reconstructions
- run single-step image diffusion inference with diffusers
- train a custom artifact-removal diffusion model

## When to choose
- you need to fix artifacts in NeRF or 3DGS renders from extreme viewpoints
- you want a fast single-step diffusion enhancer integrated with Hugging Face diffusers
- you are doing research on diffusion-based 3D reconstruction enhancement

## When to avoid
- you need a general-purpose image restoration tool unrelated to 3D reconstruction
- you lack a CUDA GPU
- you need a production-ready product with support rather than research code
- you need a permissively licensed library for commercial embedding without reviewing the custom license

## Facets
- artifact type: library
- maturity: active
- function: image-processing, machine-learning, deep-learning, computer-vision
- domain: computer-vision, graphics, artificial-intelligence, deep-learning
- platform: python
- tags: diffusion-models, gaussian-splatting, nerf, 3d-reconstruction, novel-view-synthesis, cvpr-2025, nvidia, image-restoration, gpu, linux

## Member repositories
- nv-tlabs/Difix3D (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:11.188045+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 2026-08-30T05:03:48.381082+00:00, confidence not recorded.
  - readme: https://github.com/nv-tlabs/Difix3D (fetched 2026-08-28T04:04:11.188045+00:00, sha 96ccb9259094)
  - homepage: https://research.nvidia.com/labs/toronto-ai/difix3d/ (fetched 2026-08-29T12:15:56.015516+00:00, sha c86abf12fcaf)
- Data as of 2026-08-30T08:39:29.467469+00:00.