Guardado en:
Detalles Bibliográficos
Autores principales: Luo, Yan, Aidara, Ahmadou, Lu, Jingyi, Moebel, Jeremy, Han, Kai, Wang, Mengyu
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2605.15661
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866909046379905024
author Luo, Yan
Aidara, Ahmadou
Lu, Jingyi
Moebel, Jeremy
Han, Kai
Wang, Mengyu
author_facet Luo, Yan
Aidara, Ahmadou
Lu, Jingyi
Moebel, Jeremy
Han, Kai
Wang, Mengyu
contents Classifier-free guidance (CFG) is the primary control over how strongly text semantics move a flow-based sampler, yet standard practice holds its scale fixed across the entire ODE trajectory. This is a fundamental mismatch: early steps are noise-dominated and carry weak semantic signal, while late steps commit image structure and demand stronger directional commitment; more critically, the value of any guidance strength depends on whether the guided velocity is consistent with the model's current dynamics or working against them. We propose \textit{Velocity-Adaptive Guidance Scale} (VAGS), a training-free replacement that multiplies the nominal scale by a bounded factor combining a temporal signal-level term with the cosine similarity between task-relevant velocity fields. For inversion-free editing, VAGS measures the alignment between source- and target-guided velocities, so edit strength at each step reflects local compatibility between preservation and transformation. For generation, VAGS-Gen uses the alignment between unconditional and conditional velocities as the analogous signal. Neither variant requires fine-tuning, auxiliary networks, or extra forward passes, and fixed CFG is recovered as a special case. On PIE-Bench and DIV2K for editing, and COCO17, CUB-200, and Flickr30K for generation, VAGS consistently improves structural fidelity and generation quality over fixed CFG and recent training-free guidance variants. The code is publicly available at https://github.com/Harvard-AI-and-Robotics-Lab/Velocity_Adaptive_Guidance_Scale.
format Preprint
id arxiv_https___arxiv_org_abs_2605_15661
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle VAGS: Velocity Adaptive Guidance Scale for Image Editing and Generation
Luo, Yan
Aidara, Ahmadou
Lu, Jingyi
Moebel, Jeremy
Han, Kai
Wang, Mengyu
Computer Vision and Pattern Recognition
Artificial Intelligence
Classifier-free guidance (CFG) is the primary control over how strongly text semantics move a flow-based sampler, yet standard practice holds its scale fixed across the entire ODE trajectory. This is a fundamental mismatch: early steps are noise-dominated and carry weak semantic signal, while late steps commit image structure and demand stronger directional commitment; more critically, the value of any guidance strength depends on whether the guided velocity is consistent with the model's current dynamics or working against them. We propose \textit{Velocity-Adaptive Guidance Scale} (VAGS), a training-free replacement that multiplies the nominal scale by a bounded factor combining a temporal signal-level term with the cosine similarity between task-relevant velocity fields. For inversion-free editing, VAGS measures the alignment between source- and target-guided velocities, so edit strength at each step reflects local compatibility between preservation and transformation. For generation, VAGS-Gen uses the alignment between unconditional and conditional velocities as the analogous signal. Neither variant requires fine-tuning, auxiliary networks, or extra forward passes, and fixed CFG is recovered as a special case. On PIE-Bench and DIV2K for editing, and COCO17, CUB-200, and Flickr30K for generation, VAGS consistently improves structural fidelity and generation quality over fixed CFG and recent training-free guidance variants. The code is publicly available at https://github.com/Harvard-AI-and-Robotics-Lab/Velocity_Adaptive_Guidance_Scale.
title VAGS: Velocity Adaptive Guidance Scale for Image Editing and Generation
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2605.15661