Figure 1: Our method is Training-Free, Universal, and Plug-and-Play. It significantly improves alignment and quality across SDXL, FLUX.1, Wan 2.1 (Video), and TRELLIS (3D). In text-driven content ...