Envisioning Beyond the Few: Disentangled Semantics and Primitives for Few-Shot Atypical Layout-to-Image Generation

Researchers introduced a representation-driven layout-to-image framework that separates semantic identity from local visual details.

The method uses Semantic Anchoring, Primitive Imbuing, and Conceptual Steering to reduce representation fragmentation in few-shot atypical layouts.

It improves semantic alignment and image fidelity, showing stronger 5-shot results than prior approaches.

The work targets a common failure mode in layout-to-image generation, especially for rare or unusual object arrangements.