DanceOPD: On-Policy Generative Field Distillation

• URL: http://arxiv.org/abs/2606.27377v1
• Date: Thu, 25 Jun 2026 17:59:58 GMT
• Title: DanceOPD: On-Policy Generative Field Distillation
• Abstract summary: We introduce DanceOPD, an on-policy generative field distillation framework for flow-matching models.<n>It routes each sample to one capability field, queries one low-noise student-induced state, and trains with a simple velocity MSE objective.<n> Comprehensive experiments on T2I, editing, realism-field absorption, and CFG absorption show that our approach improves multi-capability composition.
• Score: 62.31008793798276
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: Modern image generation demands a single model that unifies diverse capabilities, including text-to-image (T2I), local editing, and global editing. However, these capabilities are rarely naturally aligned and often conflict. For instance, editing tends to degrade T2I performance, while global and local editing interfere with each other. Consequently, effectively composing these capabilities has become a central challenge for image generation model training. To tackle this, we introduce DanceOPD, an on-policy generative field distillation framework for flow-matching models that routes each sample to one capability field, queries one low-noise student-induced state, and trains with a simple velocity MSE objective. With each capability source defined as a velocity field over the shared flow state space, the student learns from fields queried on its own rollout states to compose expert capabilities. This formulation also absorbs operator-defined fields such as classifier-free guidance. Comprehensive experiments on T2I, editing, realism-field absorption, and CFG absorption show that our approach improves multi-capability composition, strengthening target capabilities while preserving anchor generation quality. We believe this work establishes a practical route for generative field distillation in flow-matching models.

Related papers

• Adversarial Domain Prompt Tuning and Generation for Single Domain Generalization [105.71296914822898]
  Single domain generalization aims to learn a robust model, which could perform well on many unseen domains.<n>One of the promising directions for achieving single-domain generalization is to generate out-of-domain (OOD) training data through data augmentation or image generation.<n>This paper is the first to propose leveraging powerful pre-trained text-to-image (T2I) foundation models to create the training data.
  arXiv  Detail & Related papers  (2026-06-19T20:46:24Z)
• D-OPSD: On-Policy Self-Distillation for Continuously Tuning Step-Distilled Diffusion Models [27.90064267850009]
  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.
  arXiv  Detail & Related papers  (2026-05-06T17:59:34Z)
• Envision: Benchmarking Unified Understanding & Generation for Causal World Process Insights [41.385614383367205]
  Current models aim to transcend the limitations of single-modality representations by unifying understanding and generation.<n>Their reliance on static single-image generation in training and evaluation leads to overfitting to static pattern matching and semantic fusion.<n>We propose Envision-a causal event progression benchmark for chained text-to-multi-image generation.
  arXiv  Detail & Related papers  (2025-12-01T15:52:31Z)
• Dual Recursive Feedback on Generation and Appearance Latents for Pose-Robust Text-to-Image Diffusion [15.384896404310645]
  We propose a training-free Dual Recursive Feedback (DRF) system that properly reflects control conditions in controllable T2I models.<n>Our method produces high-quality, semantically coherent, and structurally consistent image generations.
  arXiv  Detail & Related papers  (2025-08-13T07:46:00Z)
• UniSTD: Towards Unified Spatio-Temporal Learning across Diverse Disciplines [64.84631333071728]
  We introduce bfUnistage, a unified Transformer-based framework fortemporal modeling.<n>Our work demonstrates that a task-specific vision-text can build a generalizable model fortemporal learning.<n>We also introduce a temporal module to incorporate temporal dynamics explicitly.
  arXiv  Detail & Related papers  (2025-03-26T17:33:23Z)
• DiffMoE: Dynamic Token Selection for Scalable Diffusion Transformers [86.5541501589166]
  DiffMoE is a batch-level global token pool that enables experts to access global token distributions during training.<n>It achieves state-of-the-art performance among diffusion models on ImageNet benchmark.<n>The effectiveness of our approach extends beyond class-conditional generation to more challenging tasks such as text-to-image generation.
  arXiv  Detail & Related papers  (2025-03-18T17:57:07Z)
• UniCombine: Unified Multi-Conditional Combination with Diffusion Transformer [24.159791066104358]
  We introduce a DiT-based multi-conditional controllable generative framework capable of handling any combination of conditions.<n>Specifically, we introduce a novel MMDiT Attention mechanism and incorporate a trainable LoRA module.<n>We also propose a new pipeline to construct SubjectSpatial200K, the first dataset designed for multi-conditional generative tasks.
  arXiv  Detail & Related papers  (2025-03-12T11:22:47Z)
• Towards Enhanced Image Generation Via Multi-modal Chain of Thought in Unified Generative Models [52.84391764467939]
  Unified generative models have shown remarkable performance in text and image generation.<n>We introduce Chain of Thought (CoT) into unified generative models to address the challenges of complex image generation.<n>Experiments show that FoX consistently outperforms existing unified models on various T2I benchmarks.
  arXiv  Detail & Related papers  (2025-03-03T08:36:16Z)
• Stable Flow: Vital Layers for Training-Free Image Editing [74.52248787189302]
  Diffusion models have revolutionized the field of content synthesis and editing.<n>Recent models have replaced the traditional UNet architecture with the Diffusion Transformer (DiT)<n>We propose an automatic method to identify "vital layers" within DiT, crucial for image formation.<n>Next, to enable real-image editing, we introduce an improved image inversion method for flow models.
  arXiv  Detail & Related papers  (2024-11-21T18:59:51Z)
• Divide and Conquer: Rethinking the Training Paradigm of Neural Radiance Fields [24.99410489251996]
  Neural radiance fields (NeRFs) have exhibited potential in high-fidelity views of 3D scenes. Standard training paradigm of NeRF presupposes an equal importance for each image in the training set. In this paper, we take a closer look at the implications of the current training paradigm and redesign this for more superior rendering quality.
  arXiv  Detail & Related papers  (2024-01-29T13:23:34Z)

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.