Moebius: 0.2B Lightweight Image Inpainting Framework Matches 11.9B FLUX Model
Huazhong University of Science and Technology
Moebius introduces a 0.22B parameter image inpainting model that matches or surpasses FLUX.1-Fill-Dev (11.9B parameters) through a Local-λ Mix Interaction block that summarizes spatial context and global semantic priors into fixed-size linear matrices. Adaptive multi-granularity latent-space distillation delivers a 15× inference speedup.
Why it matters
Top-voted paper on HuggingFace Daily Papers with over 100 upvotes. Demonstrates that extreme parameter efficiency (under 2% of a baseline model's size) is achievable for a demanding generative task without quality loss.
Importance: 4/5
100+ HF Daily upvotes; 50× parameter reduction vs FLUX.1-Fill-Dev at matching quality — a new efficiency frontier for diffusion inpainting.