Safe Few-Step Generation via Velocity Editing

arXiv:2606.23267 · cs.CV, cs.CY · Submitted 2026-06-22 · Read on arXiv

cs.CV, cs.CY

Submitted: 2026-06-22

Updated: 2026-10-10

Code: https://github.com/LAION-AI/CLIP-based-NSFW-Detector

Project page: https://uzn36.github.io/VESFlow

Terminology

• Flow Matching — 4× in this paper · explained in 10 episodes
• Fidelity — 1× in this paper · explained in 2 episodes
• Flow Matching Model — 1× in this paper · explained in 3 episodes
• Steering — 1× in this paper · explained in 2 episodes

Sources

• Mean Flows for One-step Generative Modeling
• Classifier-Free Diffusion Guidance
• Training-Free Safe Denoisers for Safe Use of Diffusion Models
• Safety-Guided Flow (SGF): A Unified Framework for Negative Guidance in Safe Generation
• Shielded Diffusion: Generating Novel and Diverse Images using Sparse Repellency
• EraseFlow: Learning Concept Erasure Policies via GFlowNet-Driven Alignment
• Flow Matching for Generative Modeling
• Training-Free Safe Text Embedding Guidance for Text-to-Image Diffusion Models
• Few-Step Distillation for Text-to-Image Generation: A Practical Guide
• Score-Based Generative Modeling through Stochastic Differential Equations
• Ring-A-Bell! How Reliable are Concept Removal Methods for Diffusion Models?
• Tackling the Generative Learning Trilemma with Denoising Diffusion GANs
• Semantic Surgery: Zero-Shot Concept Erasure in Diffusion Models
• SAFREE: Training-Free and Adaptive Guard for Safe Text-to-Image And Video Generation

Related papers

• Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
• AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
• Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
• MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
• TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
• A Survey on Efficient Vision-Language-Action Models