Despite significant advancements in image generation using advanced generative frameworks, cross-image integration of content and style remains a key challenge. Current generative models, while powerful, frequently depend on vague textual prompts to define styles—creating difficulties in balancing content semantics and style preservation. We propose a novel framework that utilizes customized models to learn style representations. It enhances content preservation through cross-model feature and attention modulation, leveraging the inherent semantic consistency across models. Additionally, we introduce fixed feature and adaptive attention fusion to achieve the desired balance between content and style. We further develop spatial (mask-guided localized) and temporal (multi-style compositional) multi-model combinations, enabling flexible fusion of models and styles. Extensive experiments demonstrate that our method outperforms state-of-the-art approaches in balancing content preservation and stylistic coherence.

Zero-shot text-to-video diffusion models are crafted to expand pre-trained image diffusion models to the video domain without additional training. In recent times, prevailing techniques commonly rely on existing shapes as constraints and introduce inter-frame attention to ensure texture consistency. However, such shape constraints tend to restrict the stylized geometric deformation of videos and inadvertently neglect the original texture characteristics. Furthermore, existing methods suffer from flickering and inconsistent facial expressions. In this paper, we present DiffPortraitVideo. The framework employs a diffusion model-based feature and attention injection mechanism to generate key frames, with cross-frame constraints to enforce coherence and adaptive feature fusion to ensure expression consistency. Our approach achieves high spatio-temporal and expression consistency while retaining the textual and original image properties. Extensive and comprehensive experiments are conducted to validate the efficacy of our proposed framework in generating personalized, high-quality, and coherent videos. This not only showcases the superiority of our method over existing approaches but also paves the way for further research and development in the field of text-to-video generation with enhanced personalization and quality.

We present Instant Photorealistic Neural Radiance Fields Stylization, a novel approach for multi-view image stylization for the 3D scene. Our approach models a neural radiance field based on neural graphics primitives, which use a hash table-based position encoder for position embedding. We split the position encoder into two parts, the content and style sub-branches, and train the network for normal novel view image synthesis with the content and style targets. In the inference stage, we execute AdaIN to the output features of the position encoder, with content and style voxel grid features as reference. The stylization of novel view images could be obtained with the adjusted features. Given a set of images of 3D scenes and a style target(a style image or another set of 3D scenes), our method can generate stylized novel views with a consistent appearance at various view angles in less than 10 minutes on modern GPU hardware.

Both 2D images and 3D models are vital aspects of portrait applications. Existing style transfer methods principally emphasized 2D images, neglecting the urge for 3D style transfer. We propose rendering and reconstruction based 3D portrait stylization. And we present the first geometry-aware stereoscopic image stylization. Our framework requires one content image and one style image to obtain a 3D stylization portrait. In the first step, 3D face reconstruction produces a 3D face model. We excute stereoscopic rendering to the model and reserve the images and parameters. We propose to perform the perspective transformation on one style image to match two content images. Then we use disparity loss to conduct a geometry-aware stereoscopic stylization. Using stereoscopic stylization images, we calculate the 3D stylization portrait using a stereoscopic 3D reconstruction algorithm. Expect for portraits, the framework applies to models with simple shapes. Extensive experiments demonstrate the validity and robustness of our method.