kvuong2711/fix-anything

FixAnything: 3D-Consistent Rendering Refinement via Video Generative Priors

Khiem Vuong, Deva Ramanan*, Srinivasa Narasimhan* — ECCV 2026

[arXiv] [Project Page] [Code]

FixAnything is a single generalist video model that repairs rendering artifacts from any 3D representation — 3DGS, NeRF, meshes, or sparse point clouds — by repurposing a pretrained video diffusion model with minimal modification and finetuning.

Model

This repository holds the FixAnything weights: a rank-64 LoRA on top of Wan2.1-I2V-14B-480P.

File	Description
fixanything_lora.safetensors	LoRA weights (rank 64) for the Wan2.1-I2V-14B DiT

The Wan2.1-I2V-14B-480P base model is required and is downloaded separately (see the code repository).

Usage

git clone https://github.com/kvuong2711/fix-anything.git && cd fix-anything
# ... install (see README) ...
huggingface-cli download kvuong2711/fix-anything fixanything_lora.safetensors --local-dir checkpoints
python scripts/run_inference.py --input examples/dl3dv_3dgs/input.mp4 --output_dir outputs/dl3dv_3dgs

License

Released under the Apache 2.0 License, the license of the underlying Wan2.1 model.

Citation

@inproceedings{vuong2026fixanything,
  title     = {FixAnything: 3D-Consistent Rendering Refinement via Video Generative Priors},
  author    = {Vuong, Khiem and Ramanan, Deva and Narasimhan, Srinivasa},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2026}
}