The field of generative models is rapidly advancing, with a focus on improving the efficiency, quality, and controllability of flow-based models. Recent developments have led to the proposal of novel frameworks, such as Blockwise Flow Matching and Improved Training Technique for Shortcut Models, which address limitations in existing models and achieve state-of-the-art results. The use of flow-based models has also been extended to various applications, including text-to-image generation, image editing, and autonomous driving. Noteworthy papers in this area include Blockwise Flow Matching, which improves inference efficiency and sample quality, and SplitFlow, which enables inversion-free text-to-image editing with high fidelity and diversity. Overall, the field is moving towards more efficient, flexible, and controllable generative models that can be applied to a wide range of tasks.
Advances in Flow-Based Generative Models
Sources
Blockwise Flow Matching: Improving Flow Matching Models For Efficient High-Quality Generation
Improved Training Technique for Shortcut Models
Towards a Golden Classifier-Free Guidance Path via Foresight Fixed Point Iterations
Sample By Step, Optimize By Chunk: Chunk-Level GRPO For Text-to-Image Generation
A Flow Model with Low-Rank Transformers for Incomplete Multimodal Survival Analysis
FlowOpt: Fast Optimization Through Whole Flow Processes for Training-Free Editing
MAGIC-Flow: Multiscale Adaptive Conditional Flows for Generation and Interpretable Classification
Discovering Latent Graphs with GFlowNets for Diverse Conditional Image Generation
GRPO-Guard: Mitigating Implicit Over-Optimization in Flow Matching via Regulated Clipping
Self-Attention Decomposition For Training Free Diffusion Editing
The Generation Phases of Flow Matching: a Denoising Perspective
Target-Guided Bayesian Flow Networks for Quantitatively Constrained CAD Generation
Balanced conic rectified flow
CDFlow: Building Invertible Layers with Circulant and Diagonal Matrices
RegionE: Adaptive Region-Aware Generation for Efficient Image Editing
BOLT-GAN: Bayes-Optimal Loss for Stable GAN Training
SplitFlow: Flow Decomposition for Inversion-Free Text-to-Image Editing
Beyond Imitation: Constraint-Aware Trajectory Generation with Flow Matching For End-to-End Autonomous Driving