DetailAnywhere: Fashion Detail Generation via Cross-Modal Feature Alignment Distillation

2026-07-02

AI Summary

This work addresses the challenge of generating high-fidelity, semantically consistent local fashion details from user-specified garment reference regions without relying on templates—a capability lacking in existing methods. To this end, we introduce the novel task of template-free fashion detail generation and establish FDBench, a dedicated benchmark for evaluation. We propose a multimodal diffusion Transformer-based framework that incorporates a cross-modal feature alignment distillation mechanism, leveraging a fine-tuned DINOv3 teacher model for dual-branch distillation, and further integrates a three-dimensional consistency reward model to guide reinforcement learning optimization. This design effectively bridges the semantic gap between global context and local details. Extensive experiments demonstrate that our approach significantly outperforms current open-source state-of-the-art methods on FDBench, achieving superior performance in both automatic metrics and human evaluations.

Abstract

Diffusion-based generative AI has achieved remarkable success in e-commerce applications such as virtual try-on, poster generation, and product background synthesis. However, when making online purchasing decisions for apparel, consumers also desire the freedom to examine specific detail regions of interest, such as collars, cuffs, and fabric textures, yet existing methods have not explicitly studied this setting. We therefore formalize a new, non-template task: Fashion Detail Generation with focus conditioning, and release FDBench, the first benchmark comprising 40K+ human-verified reference-detail pairs across 41 different categories. This task poses a unique semantic gap challenge: the model must bridge the correspondence between a focus marker on a product reference image and a photorealistic close-up view of the indicated region, while faithfully preserving the garment's identity, without any precise prompt. To bridge this gap, we propose Cross-modal Feature Alignment Distillation (CFAD), which leverages a fine-tuned DINOv3 teacher to align both branches of a Multimodal Diffusion Transformer in a shared semantic space via dual-branch distillation. To further improve consistency between generated details and reference images, we introduce a consistency reward model that jointly scores image pairs along three quality axes and optimizes generation via reinforcement learning. Experiments show that our model DetailAnywhere significantly outperforms all state-of-the-art opensource methods across all metrics and human evaluations.

Problem

• Fashion Detail Generation
• Cross-Modal Alignment
• Diffusion Models
• Semantic Gap
• Focus Conditioning

Innovation

• Cross-modal Feature Alignment Distillation
• Fashion Detail Generation
• Multimodal Diffusion Transformer
• Consistency Reward Model
• Focus Conditioning