D-OPSD: On-Policy Self-Distillation for Continuously Tuning Step-Distilled Diffusion Models

• URL: http://arxiv.org/abs/2605.05204v1
• Date: Wed, 06 May 2026 17:59:34 GMT
• Title: D-OPSD: On-Policy Self-Distillation for Continuously Tuning Step-Distilled Diffusion Models
• Abstract summary: We propose D-OPSD, a novel training paradigm for step-distilled diffusion models.<n>D-OPSD enables the model to learn new concept, style, etc. without sacrificing the original few-step capacity.
• Score: 27.90064267850009
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: The landscape of high-performance image generation models is currently shifting from the inefficient multi-step ones to the efficient few-step counterparts (e.g, Z-Image-Turbo and FLUX.2-klein). However, these models present significant challenges for directly continuous supervised fine-tuning. For example, applying the commonly used fine-tuning technique would compromises their inherent few-step inference capability. To address this, we propose D-OPSD, a novel training paradigm for step-distilled diffusion models that enables on-policy learning during supervised fine-tuning. We first find that the modern diffusion model where the LLM/VLM serves as the encoder can inherit its encoder's in-context capabilities. This enables us to make the training as an on-policy self-distillation process. Specifically, during training, we make the model acts as both the teacher and the student with different contexts, where the student is conditioned only on the text feature, while the teacher is conditioned on the multimodal feature of both the text prompt and the target image. Training minimizes the two predicted distributions over the student's own roll-outs. By optimized on the model's own trajectory and under it's own supervision, D-OPSD enables the model to learn new concept, style, etc. without sacrificing the original few-step capacity.

Related papers

• Learning an Image Editing Model without Image Editing Pairs [83.03646586929638]
  Recent image editing models have achieved impressive results while following natural language editing instructions.<n>They rely on supervised fine-tuning with large datasets of input-target pairs.<n>Current workarounds use synthetic training pairs that leverage the zero-shot capabilities of existing models.<n>We present a new training paradigm that eliminates the need for paired data entirely.
  arXiv  Detail & Related papers  (2025-10-16T17:59:57Z)
• TAID: Temporally Adaptive Interpolated Distillation for Efficient Knowledge Transfer in Language Models [6.8298782282181865]
  We introduce $textitTemporally Adaptive Interpolated Distillation (TAID)$, a novel knowledge distillation approach.<n>We show TAID's superior performance across various model sizes and architectures in both instruction tuning and pre-training scenarios.<n>These results demonstrate TAID's effectiveness in creating high-performing and efficient models, advancing the development of more accessible AI technologies.
  arXiv  Detail & Related papers  (2025-01-28T13:31:18Z)
• OFTSR: One-Step Flow for Image Super-Resolution with Tunable Fidelity-Realism Trade-offs [24.046764908874703]
  OFTSR is a flow-based framework for one-step image super-resolution that can produce outputs with tunable levels of fidelity and realism.<n>We demonstrate that OFTSR achieves state-of-the-art performance for one-step image super-resolution, while having the ability to flexibly tune the fidelity-realism trade-off.
  arXiv  Detail & Related papers  (2024-12-12T17:14:58Z)
• Exploring and Enhancing the Transfer of Distribution in Knowledge Distillation for Autoregressive Language Models [62.5501109475725]
  Knowledge distillation (KD) is a technique that compresses large teacher models by training smaller student models to mimic them. This paper introduces Online Knowledge Distillation (OKD), where the teacher network integrates small online modules to concurrently train with the student model. OKD achieves or exceeds the performance of leading methods in various model architectures and sizes, reducing training time by up to fourfold.
  arXiv  Detail & Related papers  (2024-09-19T07:05:26Z)
• One-Step Image Translation with Text-to-Image Models [35.0987002313882]
  We introduce a general method for adapting a single-step diffusion model to new tasks and domains through adversarial learning objectives. We consolidate various modules of the vanilla latent diffusion model into a single end-to-end generator network with small trainable weights. Our model CycleGAN-Turbo outperforms existing GAN-based and diffusion-based methods for various scene translation tasks.
  arXiv  Detail & Related papers  (2024-03-18T17:59:40Z)
• Adaptive Training Meets Progressive Scaling: Elevating Efficiency in Diffusion Models [52.1809084559048]
  We propose a novel two-stage divide-and-conquer training strategy termed TDC Training.<n>It groups timesteps based on task similarity and difficulty, assigning highly customized denoising models to each group, thereby enhancing the performance of diffusion models.<n>While two-stage training avoids the need to train each model separately, the total training cost is even lower than training a single unified denoising model.
  arXiv  Detail & Related papers  (2023-12-20T03:32:58Z)
• BOOT: Data-free Distillation of Denoising Diffusion Models with Bootstrapping [64.54271680071373]
  Diffusion models have demonstrated excellent potential for generating diverse images. Knowledge distillation has been recently proposed as a remedy that can reduce the number of inference steps to one or a few. We present a novel technique called BOOT, that overcomes limitations with an efficient data-free distillation algorithm.
  arXiv  Detail & Related papers  (2023-06-08T20:30:55Z)
• eP-ALM: Efficient Perceptual Augmentation of Language Models [70.47962271121389]
  We propose to direct effort to efficient adaptations of existing models, and propose to augment Language Models with perception. Existing approaches for adapting pretrained models for vision-language tasks still rely on several key components that hinder their efficiency. We show that by freezing more than 99% of total parameters, training only one linear projection layer, and prepending only one trainable token, our approach (dubbed eP-ALM) significantly outperforms other baselines on VQA and Captioning.
  arXiv  Detail & Related papers  (2023-03-20T19:20:34Z)
• Reinforced Multi-Teacher Selection for Knowledge Distillation [54.72886763796232]
  knowledge distillation is a popular method for model compression. Current methods assign a fixed weight to a teacher model in the whole distillation. Most of the existing methods allocate an equal weight to every teacher model. In this paper, we observe that, due to the complexity of training examples and the differences in student model capability, learning differentially from teacher models can lead to better performance of student models distilled.
  arXiv  Detail & Related papers  (2020-12-11T08:56:39Z)

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.