CharacterFactory is a groundbreaking framework that enables the sampling of new characters with consistent identities in the latent space of Generative Adversarial Networks (GANs) for diffusion models. This innovative approach addresses the limitations of current text-to-image models, which cannot directly generate images with consistent, newly coined identities.
The framework considers the word embeddings of celebrity names as ground truths for the identity-consistent generation task. It trains a GAN model to learn the mapping from a latent space to the celebrity embedding space. Additionally, CharacterFactory incorporates a context-consistent loss to ensure that the generated identity embeddings can produce identity-consistent images across various contexts.
One of the most remarkable aspects of CharacterFactory is its efficiency. The entire model requires only 10 minutes for training and can sample an infinite number of characters end-to-end during inference. Extensive experiments demonstrate the exceptional performance of CharacterFactory in character creation, particularly in terms of identity consistency and editability.
Moreover, the characters generated by CharacterFactory can be seamlessly integrated with off-the-shelf image, video, and 3D diffusion models. This compatibility opens up a wide range of possibilities for creating consistent and engaging characters across different media formats.
CharacterFactory represents a significant step forward in identity-consistent character generation. Its ability to efficiently create new characters with consistent identities has the potential to revolutionize various industries, such as gaming, animation, and virtual reality, by enabling the rapid development of diverse and compelling characters.
Using GANs for diffusion models, Character Factory aims to generate consistent characters with high accuracy. This approach enhances the reliability of character generation in various applications, ensuring stable and predictable results.
