Q-Sched: Pushing the Boundaries of Few-Step Diffusion Models with Quantization-Aware Scheduling

Text-to-image diffusion models remain computationally intensive: generating a single image typically requires dozens of passes through large transformer backbones (e.g., SDXL uses ~50 evaluations of a 2.6B-parameter model). Few-step variants reduce the step count to 2–8 but still rely on large, full-precision backbones, making inference impractical on resource-constrained platforms. Existing post-training quantization (PTQ) methods are further hampered by their dependence on full-precision calibration. We introduce Q-Sched, a scheduler-level PTQ approach that adapts the diffusion sampler while keeping the quantized weights fixed. By adjusting the few-step sampling trajectory with quantization-aware preconditioning coefficients, Q-Sched matches or surpasses full-precision quality while delivering a 4× reduction in model size and preserving a single reusable checkpoint across bit-widths. To learn these coefficients, we propose a reference-free Joint Alignment–Quality (JAQ) loss, which combines text–image compatibility with an image-quality objective for fine-grained control. JAQ requires only a handful of calibration prompts and avoids any full-precision inference during calibration. Empirically, Q-Sched yields substantial gains: a 15.5% FID improvement over the FP16 4-step Latent Consistency Model and a 16.6% improvement over the FP16 8-step Phased Consistency Model, demonstrating that quantization and few-step distillation are complementary for high-fidelity generation. A large-scale user study with 80,000+ annotations further validates these results on both FLUX.1[schnell] and SDXL-Turbo. Code: https://github.com/enyac-group/q-sched

Modern AI image generators can create impressive images from text, but they are expensive to run because they repeatedly use very large neural networks. While recent advances have reduced the number of generation steps needed, these systems still require substantial computing resources, making them difficult to deploy on phones, laptops, and other resource-constrained devices. We developed Q-Sched, a technique that makes image generation models much smaller and more efficient while maintaining high image quality. Instead of modifying the model itself, Q-Sched adjusts the image generation process to compensate for quality loss introduced by model compression. This allows a model that is four times smaller to match or even outperform the original full-size version. Q-Sched also avoids the need for expensive calibration using the original model, making deployment simpler and more practical. Across multiple state-of-the-art image generation systems, Q-Sched consistently improved image quality, and a large human study with more than 80,000 ratings confirmed that people preferred its outputs. These results show that AI image generators can become significantly more efficient without sacrificing quality, helping make advanced generative AI more accessible and affordable.