# David and Goliath: Small One-step Model Beats Large Diffusion with Score Post-training

arXiv: https://arxiv.org/abs/2410.20898 · HTML: https://mlsanity.com/paper/2410.20898/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- Placed in Diffusion and multimodal model alignment, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 73 of its 83 references that are in the map.
- First posted on arXiv 28 Oct 2024.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Diffusion and multimodal model alignment](https://mlsanity.com/alignment/diffusion-and-multimodal-model-alignment/)

## Builds on (most cited first, 30 of 73)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Denoising Diffusion Probabilistic Models](https://mlsanity.com/paper/2006.11239/) (2020)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Score-Based Generative Modeling through Stochastic Differential Equations](https://mlsanity.com/paper/2011.13456/) (2020)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Deep Unsupervised Learning using Nonequilibrium Thermodynamics](https://mlsanity.com/paper/1503.03585/) (2015)
- [Hierarchical Text-Conditional Image Generation with CLIP Latents](https://mlsanity.com/paper/2204.06125/) (2022)
- [Classifier-Free Diffusion Guidance](https://mlsanity.com/paper/2207.12598/) (2022)
- [SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis](https://mlsanity.com/paper/2307.01952/) (2023)
- [Scaling Rectified Flow Transformers for High-Resolution Image Synthesis](https://mlsanity.com/paper/2403.03206/) (2024)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [Improved Denoising Diffusion Probabilistic Models](https://mlsanity.com/paper/2102.09672/) (2021)
- [Analyzing and Improving the Image Quality of StyleGAN](https://mlsanity.com/paper/1912.04958/) (2019)
- [DreamFusion: Text-to-3D using 2D Diffusion](https://mlsanity.com/paper/2209.14988/) (2022)
- [Large Scale GAN Training for High Fidelity Natural Image Synthesis](https://mlsanity.com/paper/1809.11096/) (2018)
- [Video Diffusion Models](https://mlsanity.com/paper/2204.03458/) (2022)
- [Consistency Models](https://mlsanity.com/paper/2303.01469/) (2023)
- [WaveNet: A Generative Model for Raw Audio](https://mlsanity.com/paper/1609.03499/) (2016)
- [Scaling Autoregressive Models for Content-Rich Text-to-Image Generation](https://mlsanity.com/paper/2206.10789/) (2022)
- [One-step Diffusion with Distribution Matching Distillation](https://mlsanity.com/paper/2311.18828/) (2023)
- [Training Diffusion Models with Reinforcement Learning](https://mlsanity.com/paper/2305.13301/) (2023)
- [Pick-a-Pic: An Open Dataset of User Preferences for Text-to-Image Generation](https://mlsanity.com/paper/2305.01569/) (2023)
- [Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesis](https://mlsanity.com/paper/2306.09341/) (2023)
- [WorldSimBench: Towards Video Generation Models as World Simulators](https://mlsanity.com/paper/2410.18072/) (2024)
- [Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference](https://mlsanity.com/paper/2310.04378/) (2023)
- [Diffusion Model Alignment Using Direct Preference Optimization](https://mlsanity.com/paper/2311.12908/) (2023)
- [Adversarial Diffusion Distillation](https://mlsanity.com/paper/2311.17042/) (2023)
- [On Distillation of Guided Diffusion Models](https://mlsanity.com/paper/2210.03142/) (2022)

## Built on by (most cited first, 9 of 9)

- [Advantage Weighted Matching: Aligning RL with Pretraining in Diffusion Models](https://mlsanity.com/paper/2509.25050/) (2025)
- [Decoupled DMD: CFG Augmentation as the Spear, Distribution Matching as the Shield](https://mlsanity.com/paper/2511.22677/) (2025)
- [Adversarial Distribution Matching for Diffusion Distillation Towards Efficient Image and Video Synthesis](https://mlsanity.com/paper/2507.18569/) (2025)
- [Ultra-Fast Language Generation via Discrete Diffusion Divergence Instruct](https://mlsanity.com/paper/2509.25035/) (2025)
- [Soft-Di[M]O: Improving One-Step Discrete Image Generation with Soft Embeddings](https://mlsanity.com/paper/2509.22925/) (2025)
- [Drifting Preference Optimization for One-Step Generative Models](https://mlsanity.com/paper/2606.02521/) (2026)
- [Masked Auto-Regressive Variational Acceleration: Fast Inference Makes Practical Reinforcement Learning](https://mlsanity.com/paper/2511.15190/) (2025)
- [Unbiased Diffusion Variational Inversion via Principled Posterior Matching](https://mlsanity.com/paper/2605.25042/) (2026)
- [Efficient Fr'echet Distributional Decoder Alignment for Latent Generative Models](https://mlsanity.com/paper/2610.17466/) (2026)

## Abstract (arXiv)

We propose Diff-Instruct* (DI*), a data-efficient post-training approach for one-step text-to-image generative models to improve its human preferences without requiring image data. Our method frames alignment as online reinforcement learning from human feedback (RLHF), which optimizes the one-step model to maximize human reward functions while being regularized to be kept close to a reference diffusion process. Unlike traditional RLHF approaches, which rely on the Kullback-Leibler divergence as the regularization, we introduce a novel general score-based divergence regularization that substantially improves performance as well as post-training stability. Although the general score-based RLHF objective is intractable to optimize, we derive a strictly equivalent tractable loss function in theory that can efficiently compute its gradient for optimizations. We introduce DI*-SDXL-1step, which is a 2.6B one-step text-to-image model at a resolution of 1024× 1024, post-trained from DMD2 w.r.t SDXL. textbfOur 2.6B DI*-SDXL-1step model outperforms the 50-step 12B FLUX-dev model in ImageReward, PickScore, and CLIP score on the Parti prompts benchmark while using only 1.88% of the inference time. This result clearly shows that with proper post-training, the small one-step model is capable of beating huge multi-step diffusion models. Our model is open-sourced at this link: https://github.com/pkulwj1994/diffinstructstar. We hope our findings can contribute to human-centric machine learning techniques.