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

• URL: http://arxiv.org/abs/2410.20898v3
• Date: Thu, 05 Jun 2025 12:11:49 GMT
• Title: David and Goliath: Small One-step Model Beats Large Diffusion with Score Post-training
• Authors: Weijian Luo, Colin Zhang, Debing Zhang, Zhengyang Geng,
• Abstract summary: We propose Diff-Instruct* (DI*), a data-efficient post-training approach for one-step text-to-image generative models.<n>Our method frames alignment as online reinforcement learning from human feedback.<n>Our 2.6B emphDI*-SDXL-1step model outperforms the 50-step 12B FLUX-dev model.
• Score: 8.352666876052616
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 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 \emph{gradient} for optimizations. We introduce \emph{DI*-SDXL-1step}, which is a 2.6B one-step text-to-image model at a resolution of $1024\times 1024$, post-trained from DMD2 w.r.t SDXL. \textbf{Our 2.6B \emph{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/diff_instruct_star. We hope our findings can contribute to human-centric machine learning techniques.

Related papers

• Next-Scale Autoregressive Models are Zero-Shot Single-Image Object View Synthesizers [4.015569252776372]
  ArchonView is a method that significantly exceeds state-of-the-art methods despite being trained from scratch with 3D rendering data only and no 2D pretraining.<n>Our model also exhibits robust performance even for difficult camera poses where previous methods fail, and is several times faster in inference speed compared to diffusion.
  arXiv  Detail & Related papers  (2025-03-17T17:59:59Z)
• Accelerate High-Quality Diffusion Models with Inner Loop Feedback [50.00066451431194]
  Inner Loop Feedback (ILF) is a novel approach to accelerate diffusion models' inference.<n>ILF trains a lightweight module to predict future features in the denoising process.<n>ILF achieves strong performance for both class-to-image generation with diffusion transformer (DiT) and text-to-image generation with DiT-based PixArt-alpha and PixArt-sigma.
  arXiv  Detail & Related papers  (2025-01-22T18:59:58Z)
• When Worse is Better: Navigating the compression-generation tradeoff in visual tokenization [92.17160980120404]
  We introduce Causally Regularized Tokenization (CRT), which uses knowledge of the stage 2 generation modeling procedure to embed useful inductive biases in stage 1 latents. CRT makes stage 1 reconstruction performance worse, but makes stage 2 generation performance better by making the tokens easier to model. We match state-of-the-art discrete autoregressive ImageNet generation (2.18 FID) with less than half the tokens per image.
  arXiv  Detail & Related papers  (2024-12-20T20:32:02Z)
• Diff-Instruct++: Training One-step Text-to-image Generator Model to Align with Human Preferences [0.0]
  We introduce Diff-Instruct++ (DI++), the first, fast-converging and image data-free human preference alignment method for one-step text-to-image generators. In the experiment sections, we align both UNet-based and DiT-based one-step generators using DI++, which use the Stable Diffusion 1.5 and the PixelArt-$alpha$ as the reference diffusion processes. The resulting DiT-based one-step text-to-image model achieves a strong Aesthetic Score of 6.19 and an Image Reward of 1.24 on the validation prompt dataset
  arXiv  Detail & Related papers  (2024-10-24T16:17:18Z)
• Truncated Consistency Models [57.50243901368328]
  Training consistency models requires learning to map all intermediate points along PF ODE trajectories to their corresponding endpoints.<n>We empirically find that this training paradigm limits the one-step generation performance of consistency models.<n>We propose a new parameterization of the consistency function and a two-stage training procedure that prevents the truncated-time training from collapsing to a trivial solution.
  arXiv  Detail & Related papers  (2024-10-18T22:38:08Z)
• Stabilize the Latent Space for Image Autoregressive Modeling: A Unified Perspective [52.778766190479374]
  Latent-based image generative models have achieved notable success in image generation tasks. Despite sharing the same latent space, autoregressive models significantly lag behind LDMs and MIMs in image generation. We propose a simple but effective discrete image tokenizer to stabilize the latent space for image generative modeling.
  arXiv  Detail & Related papers  (2024-10-16T12:13:17Z)
• Tuning Timestep-Distilled Diffusion Model Using Pairwise Sample Optimization [97.35427957922714]
  We present an algorithm named pairwise sample optimization (PSO), which enables the direct fine-tuning of an arbitrary timestep-distilled diffusion model. PSO introduces additional reference images sampled from the current time-step distilled model, and increases the relative likelihood margin between the training images and reference images. We show that PSO can directly adapt distilled models to human-preferred generation with both offline and online-generated pairwise preference image data.
  arXiv  Detail & Related papers  (2024-10-04T07:05:16Z)
• Derivative-Free Guidance in Continuous and Discrete Diffusion Models with Soft Value-Based Decoding [84.3224556294803]
  Diffusion models excel at capturing the natural design spaces of images, molecules, DNA, RNA, and protein sequences. We aim to optimize downstream reward functions while preserving the naturalness of these design spaces. Our algorithm integrates soft value functions, which looks ahead to how intermediate noisy states lead to high rewards in the future.
  arXiv  Detail & Related papers  (2024-08-15T16:47:59Z)
• Plug-and-Play Diffusion Distillation [14.359953671470242]
  We propose a new distillation approach for guided diffusion models. An external lightweight guide model is trained while the original text-to-image model remains frozen. We show that our method reduces the inference of classifier-free guided latent-space diffusion models by almost half.
  arXiv  Detail & Related papers  (2024-06-04T04:22:47Z)
• Deep Reward Supervisions for Tuning Text-to-Image Diffusion Models [42.28905346604424]
  Deep Reward Tuning (DRTune) is an algorithm that supervises the final output image of a text-to-image diffusion model. DRTune consistently outperforms other algorithms, particularly for low-level control signals.
  arXiv  Detail & Related papers  (2024-05-01T15:26:14Z)
• You Only Sample Once: Taming One-Step Text-to-Image Synthesis by Self-Cooperative Diffusion GANs [13.133574069588896]
  YOSO is a novel generative model designed for rapid, scalable, and high-fidelity one-step image synthesis with high training stability and mode coverage.<n>We show that our method can serve as a one-step generation model training from scratch with competitive performance.<n>In particular, we show that the YOSO-PixArt-$alpha$ can generate images in one step trained on 512 resolution, with the capability of adapting to 1024 resolution without extra explicit training, requiring only 10 A800 days for fine-tuning.
  arXiv  Detail & Related papers  (2024-03-19T17:34:27Z)
• Score-Guided Diffusion for 3D Human Recovery [10.562998991986102]
  We present Score-Guided Human Mesh Recovery (ScoreHMR), an approach for solving inverse problems for 3D human pose and shape reconstruction. ScoreHMR mimics model fitting approaches, but alignment with the image observation is achieved through score guidance in the latent space of a diffusion model. We evaluate our approach on three settings/applications: (i) single-frame model fitting; (ii) reconstruction from multiple uncalibrated views; (iii) reconstructing humans in video sequences.
  arXiv  Detail & Related papers  (2024-03-14T17:56:14Z)
• Self-Play Fine-Tuning of Diffusion Models for Text-to-Image Generation [59.184980778643464]
  Fine-tuning Diffusion Models remains an underexplored frontier in generative artificial intelligence (GenAI) In this paper, we introduce an innovative technique called self-play fine-tuning for diffusion models (SPIN-Diffusion) Our approach offers an alternative to conventional supervised fine-tuning and RL strategies, significantly improving both model performance and alignment.
  arXiv  Detail & Related papers  (2024-02-15T18:59:18Z)
• One-step Diffusion with Distribution Matching Distillation [54.723565605974294]
  We introduce Distribution Matching Distillation (DMD), a procedure to transform a diffusion model into a one-step image generator. We enforce the one-step image generator match the diffusion model at distribution level, by minimizing an approximate KL divergence. Our method outperforms all published few-step diffusion approaches, reaching 2.62 FID on ImageNet 64x64 and 11.49 FID on zero-shot COCO-30k.
  arXiv  Detail & Related papers  (2023-11-30T18:59:20Z)
• Reinforcement Learning from Diffusion Feedback: Q* for Image Search [2.5835347022640254]
  We present two models for image generation using model-agnostic learning. RLDF is a singular approach for visual imitation through prior-preserving reward function guidance. It generates high-quality images over varied domains showcasing class-consistency and strong visual diversity.
  arXiv  Detail & Related papers  (2023-11-27T09:20:12Z)
• ACT-Diffusion: Efficient Adversarial Consistency Training for One-step Diffusion Models [59.90959789767886]
  We show that optimizing consistency training loss minimizes the Wasserstein distance between target and generated distributions. By incorporating a discriminator into the consistency training framework, our method achieves improved FID scores on CIFAR10 and ImageNet 64$times$64 and LSUN Cat 256$times$256 datasets.
  arXiv  Detail & Related papers  (2023-11-23T16:49:06Z)
• Diffusion Model Alignment Using Direct Preference Optimization [103.2238655827797]
  Diffusion-DPO is a method to align diffusion models to human preferences by directly optimizing on human comparison data. We fine-tune the base model of the state-of-the-art Stable Diffusion XL (SDXL)-1.0 model with Diffusion-DPO. We also develop a variant that uses AI feedback and has comparable performance to training on human preferences.
  arXiv  Detail & Related papers  (2023-11-21T15:24:05Z)
• Masked Diffusion Models Are Fast Distribution Learners [32.485235866596064]
  Diffusion models are commonly trained to learn all fine-grained visual information from scratch. We show that it suffices to train a strong diffusion model by first pre-training the model to learn some primer distribution. Then the pre-trained model can be fine-tuned for various generation tasks efficiently.
  arXiv  Detail & Related papers  (2023-06-20T08:02:59Z)
• DPOK: Reinforcement Learning for Fine-tuning Text-to-Image Diffusion Models [97.31200133440308]
  We propose using online reinforcement learning to fine-tune text-to-image models. We focus on diffusion models, defining the fine-tuning task as an RL problem. Our approach, coined DPOK, integrates policy optimization with KL regularization.
  arXiv  Detail & Related papers  (2023-05-25T17:35:38Z)
• Human Preference Score: Better Aligning Text-to-Image Models with Human Preference [41.270068272447055]
  We collect a dataset of human choices on generated images from the Stable Foundation Discord channel. Our experiments demonstrate that current evaluation metrics for generative models do not correlate well with human choices. We propose a simple yet effective method to adapt Stable Diffusion to better align with human preferences.
  arXiv  Detail & Related papers  (2023-03-25T10:09:03Z)
• On Distillation of Guided Diffusion Models [94.95228078141626]
  We propose an approach to distilling classifier-free guided diffusion models into models that are fast to sample from. For standard diffusion models trained on the pixelspace, our approach is able to generate images visually comparable to that of the original model. For diffusion models trained on the latent-space (e.g., Stable Diffusion), our approach is able to generate high-fidelity images using as few as 1 to 4 denoising steps.
  arXiv  Detail & Related papers  (2022-10-06T18:03:56Z)
• Score-Based Generative Modeling through Stochastic Differential Equations [114.39209003111723]
  We present a differential equation that transforms a complex data distribution to a known prior distribution by injecting noise. A corresponding reverse-time SDE transforms the prior distribution back into the data distribution by slowly removing the noise. By leveraging advances in score-based generative modeling, we can accurately estimate these scores with neural networks. We demonstrate high fidelity generation of 1024 x 1024 images for the first time from a score-based generative model.
  arXiv  Detail & Related papers  (2020-11-26T19:39:10Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.