Distribution coverage tradeoffs revealed in few-step generative diffusion models
Visualizing Distribution Coverage in Generative Diffusion Models
Machine Learning
Summary
Generating images or videos with diffusion models often takes many steps, so people create faster versions called distilled models. The authors found that while these faster models seem to produce good single results, they can lose variety in outputs when producing many samples. They used a new way to measure this loss called pass@k, showing that some training methods keep broad coverage of possible outputs better than others. This helps explain why some distilled models might look better on single tries but offer less diversity overall.
What this means in practice
- •For machine learning engineers: Improve design of fast image generation models by choosing training methods that preserve output diversity at scale.
- •For video production teams: Develop efficient video generation tools that maintain variety in outputs while reducing generation time.
Authors
Yifei Wang, Xiaoyu Wu, Tsu-Jui Fu, Chen Chen, Liang-Chieh Chen, Zhe Gan, Chen Wei
Abstract
Diffusion distillation is widely adopted to accelerate sampling, and the resulting few-step models are broadly believed to match or even surpass their multi-step teachers in generation. However, standard evaluations such as GenEval2 typically draw only one sample per prompt, so improved scores may fail to reveal losses in distribution coverage. We therefore revisit whether distilled models truly match their teachers beyond single-draw performance using \textbf{pass@$\mathbf{k}$}, which measures the probability that at least one of $k$ independent samples satisfies a quality criterion. At $k{=}1$, pass@$k$ reduces to standard single-draw evaluation. As $k$ grows, the curve reveals whether additional draws find genuinely different successes or merely revisit the same modes, directly exposing how broadly a model covers the space of valid outputs. We first show that classifier-free guidance (CFG), whose quality--coverage tradeoff is well established, is the clearest case: higher guidance improves pass@$1$, but its advantage shrinks and reverses at larger $k$. Applying pass@$k$ to few-step distilled models, we find the same tradeoff splits along training objectives: distribution-matching objectives concentrate the student's output distribution, boosting early-hit rates while eroding large-budget coverage, whereas consistency and trajectory-based objectives better preserve the teacher's coverage even at large $k$. We further show that this tradeoff extends to few-step causal video generation. Our findings reveal a previously overlooked cost of diffusion distillation: across both image and video generation, the choice of training objective fundamentally determines whether a few-step model inherits its teacher's distribution coverage or trades it away for single-draw quality.