BudgetAny/Budget3D-splats

Budget3D splats

🌐 Project page  |  💻 Code  |  🤗 Budget3D-Idol · Budget3D-Idol-Turbo  |  📄 Tech report (coming soon) · BudgetPix paper

The 3D Gaussian splats behind the interactive viewer on the Budget3D page: eight objects, each generated by Budget3D-Idol (50 steps) and by Budget3D-Idol-Turbo (5 steps) at 100, 90, 80, 70, 60, 50 and 40% of Idol's 8,192 latent tokens, from the same input image and the same noise at every budget.

Unofficial, open-source implementation. These splats are part of BudgetAny, a community reimplementation and extension of the compute-adaptive tokenization idea of BudgetPix: Compute-Adaptive Tokenization for Pixel-Space Image Diffusion (Kara et al., 2026). They are not an official release of that paper, and BudgetAny is not a company or a lab. The paper covers pixel-space image diffusion only; applying the idea to image-to-3D is an extension made here. Thanks to the BudgetPix authors for the method.

Files

path	format	per file	for
ply/{model}/{object}/b{budget}.ply	standard 3D Gaussian Splatting PLY, 262,144 Gaussians, degree-0 colour	17.8 MB	any splat viewer or tool
spz/{model}/{object}/b{budget}.spz	SPZ (degree-0 colour)	about 2.7 MB	the web viewer

{model} is idol or turbo; {object} is one of iris-scooter, nb-castle-iso, fig-cowboy, fig-viking, toy-train, toy-submarine, nb-boombox and fur-egg-chair; {budget} is 100, 090, 080, 070, 060, 050 or 040.

The PLY files are Idol's own output format, with y pointing down (most viewers turn such files 180 degrees about x). The SPZ files are already upright (y up) and leave out the Gaussians below 1/255 opacity, the same ones the page's renders leave out.

How they were made: the latents kept by the release galleries (the same samples as the page's videos and frames) were decoded with Idol's unchanged decoders, with the decoder's anchor sampling seeded the same way at every budget, and the SPZ files were written with Spark's encoder (transcodeSpz, Spark 2.4.0).

Citation

Tech report coming soon. Until then, please cite the BudgetPix paper:

@article{kara2026budgetpix,
  title={BudgetPix: Compute-Adaptive Tokenization for Pixel-Space Image Diffusion},
  author={Ozgur Kara and Yujia Chen and Daniel Watson and David Forsyth and James Matthew Rehg and Wen-Sheng Chu and Du Tran},
  journal={arXiv preprint arXiv:2610.12307},
  year={2026}
}

Built on Idol and Idol-turbo by Sperid Labs.

We acknowledge the compute support of the Health Care Engineering Systems Center within the Coordinated Science Lab at UIUC.

License

Apache-2.0, like Budget3D-Idol, Budget3D-Idol-Turbo and speridlabs/idol, which these splats come from.