CustomNet_云技派,专注ai人工智能! 154

CustomNet is designed to better integrate images of specified objects into newly generated images while preserving the original object's style and texture details. This technology brings a glimmer of hope to product image fusion.

In the text-to-image generation process, incorporating customized objects into image generation is a highly attractive feature. However, existing optimization-based and encoder-based approaches have several shortcomings, such as lengthy optimization time, insufficient identity preservation, and the widespread copy-paste effect.

To overcome these limitations, we introduce CustomNet, a novel object customization method that explicitly incorporates 3D novel view synthesis capabilities into the object customization pipeline. This integration helps adjust spatial positional relationships and viewpoints, producing diverse outputs while effectively preserving object features.

In addition, we introduce subtle designs that achieve position control and flexible background control through text descriptions or specific user-customized images, overcoming the limitations of existing 3D novel view synthesis methods. We further leverage a dataset construction pipeline to better handle real-world objects and complex backgrounds.

With these designs, our method achieves zero-shot object customization without test-time optimization while providing control over position, viewpoint, and background. As a result, our CustomNet ensures enhanced identity preservation and generates diverse, harmonious outputs.

CustomNet can simultaneously control viewpoint, position, and background within a unified framework, enabling harmonious customized image generation while effectively preserving object features and texture details. Background generation can be controlled through text descriptions (the "Generation" branch) or by providing specific user-customized images (the "Composition" branch).