Reproduce “SD-$π$XL: Generating Low-Resolution Quantized Imagery via Score Distillation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Alexandre Binninger, Olga Sorkine-Hornung

SD-πXL creates palette-constrained pixel art by optimizing per-pixel palette choices with a pretrained diffusion model. Gumbel-softmax lowers per-pixel entropy; argmax produces readable outputs and enforces palette membership. In a 56-person study, its K-means variant received 36.2% of first-place semantic rankings versus 22.0% for VectorFusion, but preferences varied by criterion. 6,000 steps took 1.5 hours on an RTX 4090.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.