OmniStyle-INR: Universal and Multimodal Style Transfer for INRs

2026-07-17

AI Summary

Existing Gaussian-based unified representations suffer from inefficiency in handling dense, continuous data such as images and videos, limiting their capacity to support high-quality multimodal style transfer. This work proposes the first general-purpose framework that adopts implicit neural representations (INRs) as a unified domain, enabling high-fidelity style transfer across 2D, 3D, video, and even 4D modalities through joint guidance from textual prompts and reference images. By transcending the limitations of conventional representations in expressiveness and generalization, the method achieves superior performance in style transfer quality, data compression, and super-resolution across diverse visual modalities, while seamlessly integrating with mainstream generative models.

Abstract

Style transfer remains a fundamental and highly important task across various data modalities, enabling creative manipulation conditioned by both reference images and textual descriptions. Recently, methods utilizing Gaussian Splatting have emerged as a unified representation for 2D images, video, 3D scenes, and 4D dynamics. However, representing videos and 2D images with Gaussian Splatting is structurally sub-optimal for dense continuous domains. The number of required Gaussians often approaches the total number of pixels, raising questions about the actual utility of such a representation for these specific modalities. In contrast, Implicit Neural Representations have established themselves as a much more popular and natural choice across all these data domains. Implicit Neural Representations naturally provide significant advantages, including data compression, inherent capabilities for super resolution, and seamless integration with deep generative models. To this end, we introduce OmniStyle-INR, a novel framework that leverages network-based continuous representations as a truly universal domain. Our approach successfully performs high-quality style transfer across all visual modalities, guided seamlessly by both text prompts and visual exemplars.

Problem

• Style Transfer
• Implicit Neural Representations
• Multimodal
• Gaussian Splatting
• Universal Representation

Innovation

• Implicit Neural Representations
• Multimodal Style Transfer
• Universal Representation
• Text-to-Image Style Guidance
• Continuous Domain Modeling