tl;dr: Moebius is a 0.22B-parameter image inpainting model that matches or beats the 11.9B FLUX.1-Fill-Dev across six benchmarks while running over 15× faster (26ms/step). It achieves this via two innovations: an LλMI block that compresses self- and cross-attention into fixed-size linear matrices to avoid quadratic cost, and a latent-space multi-granularity distillation strategy that transfers capacity from a larger teacher model (PixelHacker) using gradient-norm-adaptive loss weighting.