Reproduce “Human-aware Design Generation: Adding 3D Humans into Graphic Designs” 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

Zijin Hou, Ying Cao

POFRAY introduces design-conditioned 3D human adding, a task that predicts human pose, camera-relative framing, and placement within an existing layered graphic design. On the 3DHumanInDesign benchmark of about 10,000 posters, covers, and layouts, it outperformed FlexDM and an MLLM baseline, reducing MPJPE to 385.12 and achieving 6.52 HD-Align; professional designers preferred its outputs for human–design alignment in 61.1% of comparisons.

Key insights

  • Human-aware composition benefits from explicitly separating pose, framing, and layout rather than generating a complete 2D human image and fitting it afterward.
  • The 3D representation provides structured control over joint configuration and camera-relative placement, while preserving compatibility with editable layered-design workflows.
  • Conditioning pose generation on the surrounding design helps produce body orientations and shot scales that follow visual relationships in the composition, including multi-person arrangements.

Summary by alphaXiv.

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