MAMBO-G: Magnitude-Aware Mitigation for Boosted Guidance

High-fidelity text-to-image and text-to-video generation typically relies on Classifier-Free Guidance (CFG), but achieving optimal results often demands computationally expensive sampling schedules. In this work, we propose MAMBO-G, a training-free acceleration framework that significantly reduces computational cost by dynamically optimizing guidance magnitudes. We observe that standard CFG schedules are inefficient, applying disproportionately large updates in early steps that hinder convergence speed. MAMBO-G mitigates this by modulating the guidance scale based on the update-to-prediction magnitude ratio, effectively stabilizing the trajectory and enabling rapid convergence. This efficiency is particularly vital for resource-intensive tasks like video generation. Our method serves as a universal plug-and-play accelerator, achieving up to 3$\times$ speedup on Stable Diffusion v3.5 (SD3.5) and 4$\times$ on Lumina. Most notably, MAMBO-G accelerates the 14B-parameter Wan2.1 video model by 2$\times$ while preserving visual fidelity, offering a practical solution for efficient large-scale video synthesis. Our implementation follows a mainstream open-source diffusion framework and is plug-and-play with existing pipelines.

Modern AI systems can generate high-quality images and videos from text prompts, but doing so often requires many repeated computation steps, which makes the process slow and expensive. This paper introduces MAMBO-G, a method that makes this generation process faster without requiring the model to be retrained. The main idea is to adjust how strongly the text prompt guides the generation at each step, instead of using a fixed or inefficient guidance pattern throughout the process. By making these adjustments automatically, MAMBO-G helps the generation process reach good results in fewer steps while keeping image and video quality high. This is especially useful for video generation, where computation costs are much larger than for single images. In our experiments, MAMBO-G substantially speeds up several popular image and video generation systems, including a large video model, while preserving visual quality.