RGBX-Next: Towards Realistic Generative Rendering from G-Buffers

Zheng Zeng, Marco Salvi, Lifan Wu, Jan Novák, Daqi Lin, Saeed Hadadan, Yichen Sheng, Robert Pottorff, Shiqiu Liu, Ravi Ramamoorthi, Lingqi Yan, Miloš Hašan

arXiv preprint, 2026

arxiv

What is RGBX-Next?

RGBX-Next extends RGB↔X to images, videos, and streams. Our unified framework learns both directions: estimating G-buffers from RGB inputs (RGB→X), and generating realistic RGB outputs conditioned on G-buffers (X→RGB).

We present a recipe for adapting diffusion transformer (DiT) models to forward and inverse rendering. Training with real video data improves rendering realism, while flexible G-buffer conditioning balances explicit control with generative freedom. Streaming extensions support coherent forward and inverse rendering over long sequences.

Cite

@misc{zeng2026rgbxnext,
  title = {{RGBX-Next}: Towards Realistic Generative Rendering from {G-Buffers}},
  author = {Zeng, Zheng and Salvi, Marco and Wu, Lifan and Novák, Jan and Lin, Daqi and Hadadan, Saeed and Sheng, Yichen and Pottorff, Robert and Liu, Shiqiu and Ramamoorthi, Ravi and Yan, Lingqi and Hašan, Miloš},
  year = {2026},
  eprint = {2608.13929},
  archivePrefix = {arXiv},
  primaryClass = {cs.CV},
  url = {https://arxiv.org/abs/2608.13929}
}