The recent advancements in diffusion models have significantly enhanced the capabilities of image generation and editing. Researchers are focusing on improving the efficiency and quality of these models, with notable innovations in model customization, personalization, and adversarial techniques. Key areas of development include mitigating unintended alterations during model adaptation, enhancing the control over foreground objects in generated images, and dynamically adjusting guidance during the generation process. Additionally, there is a growing interest in the theoretical underpinnings of diffusion models, such as the analysis of Wasserstein convergence and the exploration of optimal control methods. These developments not only improve the performance of existing models but also open new avenues for applications in various fields, including art generation, style transfer, and personalized content creation. Notably, the introduction of novel frameworks like Group Diffusion Transformers and the exploration of zero-shot style-specific image variations highlight the potential for more scalable and versatile generative models. These advancements collectively push the boundaries of what is possible with diffusion models, making them more robust, efficient, and adaptable to a wide range of applications.
Advancing Diffusion Models: Efficiency, Control, and Versatility
Sources
Assessing Open-world Forgetting in Generative Image Model Customization
Personalized Image Generation with Large Multimodal Models
HYPNOS : Highly Precise Foreground-focused Diffusion Finetuning for Inanimate Objects
Dynamic Negative Guidance of Diffusion Models
Parallel Backpropagation for Inverse of a Convolution with Application to Normalizing Flows
Mitigating Embedding Collapse in Diffusion Models for Categorical Data
Straightness of Rectified Flow: A Theoretical Insight into Wasserstein Convergence
Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step
Truncated Consistency Models
Group Diffusion Transformers are Unsupervised Multitask Learners
DiffuseST: Unleashing the Capability of the Diffusion Model for Style Transfer
ConSinger: Efficient High-Fidelity Singing Voice Generation with Minimal Steps
Erasing Undesirable Concepts in Diffusion Models with Adversarial Preservation
DeepIcon: A Hierarchical Network for Layer-wise Icon Vectorization
Learning to Synthesize Graphics Programs for Geometric Artworks
Distribution Learning with Valid Outputs Beyond the Worst-Case
AttentionPainter: An Efficient and Adaptive Stroke Predictor for Scene Painting
Conjuring Semantic Similarity
Annotation-Free MIDI-to-Audio Synthesis via Concatenative Synthesis and Generative Refinement
One-Step Diffusion Distillation through Score Implicit Matching
MPDS: A Movie Posters Dataset for Image Generation with Diffusion Model
DiP-GO: A Diffusion Pruner via Few-step Gradient Optimization
How to Continually Adapt Text-to-Image Diffusion Models for Flexible Customization?
Lower Bounds for Convexity Testing
Scalable Ranked Preference Optimization for Text-to-Image Generation
Training Free Guided Flow Matching with Optimal Control
Beyond Color and Lines: Zero-Shot Style-Specific Image Variations with Coordinated Semantics
Wavetable Synthesis Using CVAE for Timbre Control Based on Semantic Label
Rectified Diffusion Guidance for Conditional Generation
Schedule Your Edit: A Simple yet Effective Diffusion Noise Schedule for Image Editing
Fast constrained sampling in pre-trained diffusion models
Towards Visual Text Design Transfer Across Languages
Diff-Instruct++: Training One-step Text-to-image Generator Model to Align with Human Preferences
Testing Support Size More Efficiently Than Learning Histograms
Stable Consistency Tuning: Understanding and Improving Consistency Models