Moebius: 0.2B Lightweight Image Inpainting Framework with 10B-Level Performance
Abstract
A lightweight image inpainting framework achieves high-fidelity results with significantly reduced parameters and inference time through novel local-global interaction blocks and adaptive distillation strategies.
While 10B-level industrial foundation models have pushed the boundaries of image inpainting, their prohibitive computational costs severely hinder practical deployment. Constructing a highly optimized task-specific specialist offers a promising solution; however, extreme structural compression inevitably triggers a severe representation bottleneck. To conquer this, we propose Moebius, a highly efficient lightweight inpainting framework. We systematically reconstruct the diffusion backbone by introducing the Local-λ Mix Interaction (LλMI) block. Comprising Local-λ and Interactive-λ modules, it elegantly summarizes spatial contexts and global semantic priors into fixed-size linear matrices, preserving complex latent interactions while drastically shedding parameters. Furthermore, to unlock the full representational capacity of this highly compact architecture, we synergistically pair it with an adaptive multi-granularity distillation strategy. Operating strictly within the latent space to avoid expensive pixel-space decoding, this strategy dynamically balances multiple gradient-based losses to achieve high-fidelity alignment. Extensive experiments across natural and portrait benchmarks demonstrate that this optimal synergy enables Moebius to rival or even surpass the generation quality of the 10B-level industrial generalist FLUX.1-Fill-Dev. Remarkably, Moebius achieves this using less than 2\% of the parameters (0.22B vs. 11.9B) while delivering a >15times acceleration in total inference time, setting a new efficiency standard for high-fidelity inpainting. Project page at https://hustvl.github.io/Moebius.
Community
Moebius is our latest AI Image Inpainting endeavor, serving as a direct continuation of our previous work, PixelHacker. Named after the concepts of "infinity" and "master painter," Moebius embodies our vision: maintaining exceptional generation quality under highly constrained computational resources while pushing the efficiency of image inpainting to its limits as much as possible.
Under the iron grip of the Scaling Law, AI research has long devolved into a grueling arms race of burning capital, compute, and data. Consequently, the academic community finds it increasingly difficult to keep pace with the ever-expanding model scales driven by the tech industry.
"But is this brute-force scaling truly the only path forward?"
Using general-purpose image inpainting as our strategic entry point, we challenge the "scale-at-all-costs" path dependency dictated by the Scaling Law narrative. Through the synergistic optimization of architectural design and knowledge distillation, Moebius achieves a remarkably compact footprint of just 0.22B parameters. It liberates high-quality image inpainting from the heavy-compute narrative of 10B+ foundation models: Across six comprehensive benchmarks spanning both natural and portrait scenes, Moebius performs on par with, and in certain scenarios surpasses, the inpainting quality of 10B+ industrial state-of-the-art (SOTA) generalist models like FLUX.1-Fill-Dev, while delivering a massive >15× inference acceleration.
💡 The core insight of Moebius can be summarized in a single equation:
Synergy × (Architecture + Distillation) = Shattering the "Impossible Triangle" of Low Parameters, Fast Inference, and High Quality
🌟 Highlights
- 📉 Extreme Parametric Efficiency (< 2%): Moebius operates with a mere 0.22B (226M) parameters, which represents less than 2% of the size of the colossal industrial giant FLUX.1-Fill-Dev (11.9B). It shatters the heavy-compute narrative, making high-quality inpainting accessible on consumer-grade and edge devices.
- ⚡ 15× Inference Speedup (26ms/step): Achieves a blistering inference latency of only 26.01 ms per step on a single GPU. Combined with optimized sampling steps, Moebius delivers an overall >15× total runtime acceleration compared to 10B-level models.
- 🏆 10B-Level Inpainting Quality (on-par-with/surpass FLUX.1-Fill-Dev across 6 benchmarks): Size contraction does not mean representation degradation. Through the synergistic optimization of architecture and distillation, Moebius performs on par with, and in certain scenarios (such as complex textures and facial plausibility), surpasses 10B-level state-of-the-art (SOTA) generalist models (FLUX.1-Fill-Dev, SD3.5 Large-Inpainting) across 6 comprehensive benchmarks spanning both natural scenes (Places2) and portrait scenes (CelebA-HQ, FFHQ).
- 💡 Synergistic Core Innovations:
- Architecture Design (LλMI Block): Reformulates both self- and cross-attention by condensing spatial context and global semantic priors into fixed-size linear matrices, bypassing quadratic computational overhead.
- Adaptive Multi-Granularity Distillation Strategy: Transfers the representational capacity from our PixelHacker
(teacher) strictly within the latent space (avoiding expensive pixel-space decoding). It bridges the giant capacity gap by aligning multi-granularity supervision—ranging from microscopic intermediate features to macroscopic diffusion trajectories—while dynamically balancing training via a gradient norm adaptive loss weighting mechanism. - Optimal Synergistic Balancing: Systematically explores the mutual constraint and upper bound between compact structure and distillation. By mapping this architecture-distillation synergy frontier, we ensure our 0.22B Moebius (student) absorbs the maximum semantic reasoning of PixelHacker
(teacher) without triggering representation saturation. - 🚀 Task-Specific Specialist over Bloated Generalists: Rather than blindly scaling up, Moebius answers a fundamental question: Can a model be smarter, lighter, and faster when the task is explicitly defined? It serves as a highly optimized specialist that liberates real-world image inpainting and AI object removal from parameter bloat.
TL;DR
On-par-with/surpass 10B-level industrial SOTA generalist (FLUX.1-Fill-Dev) on 6 benchmarks across natural and portrait scenes & Only 2% (0.2B) parameters, and inference 15× faster
Get this paper in your agent:
hf papers read 2606.19195 Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper
Collections including this paper 0
No Collection including this paper