# Identity Preservation

InfiniteYou (InfU) is a powerful diffusion-transformer-based framework designed for flexible image reconstruction while preserving user identity. By introducing identity features and employing a multi-stage training strategy, it significantly improves the quality and aesthetics of image generation while enhancing text-to-image alignment. This technology is important for improving the similarity and aesthetics of image generation and is suitable for various image generation tasks.

ColorFlow is a model designed for coloring image sequences, with a particular focus on preserving the identity of characters and objects during the coloring process. Utilizing contextual information, it accurately generates colors for different elements (such as hair and clothing) in black-and-white image sequences based on a pool of reference images, ensuring consistency with the color references. Through a three-phase diffusion model framework, ColorFlow introduces a novel retrieval-augmented coloring workflow that achieves relevant image coloring without the need for fine-tuning each identity or extracting explicit identity embeddings. Its main advantages include high-quality coloring effects while retaining identity information, which is of significant market value for coloring cartoon or comic series.

StableAnimator is the first end-to-end identity-preserving video diffusion framework that synthesizes high-quality videos without the need for post-processing. This technology ensures identity consistency through conditional synthesis based on reference images and a series of poses. Its main advantage is that it does not rely on third-party tools, making it suitable for users who need high-quality portrait animations.

Diffusion Self-Distillation is a self-distillation technique based on diffusion models for zero-shot custom image generation. This technology allows artists and users to generate their own datasets via a pre-trained text-to-image model without requiring large paired datasets, enabling them to fine-tune the model for image-to-image tasks conditioned on text and images. This approach surpasses existing zero-shot methods in maintaining performance on identity generation tasks and can rival instance-based tuning techniques without the need for optimization during testing.

ConsisID is a frequency decomposition-based identity-preserving text-to-video generation model that generates high-fidelity videos consistent with the input textual descriptions using identity control signals in the frequency domain. This model does not require tedious fine-tuning for different cases and is capable of maintaining consistency in character identity within the generated videos. The introduction of ConsisID advances video generation technology, particularly in terms of streamlined processes and frequency-aware identity preservation control schemes.

DisEnvisioner is an advanced image generation technology that creates customized images by separating and enhancing thematic features, eliminating the need for tedious adjustments or reliance on multiple reference images. This technology effectively distinguishes and enhances thematic features while filtering out irrelevant attributes, achieving exceptional personalization in terms of editability and identity preservation. The research basis of DisEnvisioner stems from the current demand in the field of image generation for extracting thematic features from visual cues, tackling challenges faced by existing technologies through innovative approaches.

StableIdentity is a latest advancement based on large pre-trained text-to-image models, capable of achieving high-quality, human-centric generation. Unlike existing methods, StableIdentity ensures the stable retention of identity and flexible editability, even when training is conducted using only one facial image of each subject. It utilizes a facial encoder and identity prior to encode the input face, then projects the facial representation into an editable prior space. By combining identity priors and editability priors, the learned identity can be injected into various contexts. Additionally, StableIdentity incorporates a masked two-stage diffusion loss to enhance pixel-level perception of the input face and maintain the diversity of generation. Extensive experiments demonstrate that StableIdentity outperforms previous customization methods. The learned identity can also be flexibly combined with existing modules like ControlNet. Notably, we are the first to directly inject identities learned from a single image into video/3D generation without fine-tuning. We believe that StableIdentity is an important step towards unifying image, video, and 3D customization generation models.

InstantID is a solution based on powerful diffusion models that enables personalized image processing of a single face image across various styles while ensuring high fidelity. We designed a novel IdentityNet that, through the application of strong semantic and weak spatial constraints, integrates face and landmark images with textual prompts to guide image generation. InstantID performs remarkably well in practical applications and can seamlessly integrate with popular pre-trained text-to-image diffusion models (such as SD1.5 and SDXL) as an adaptable plugin. Our code and pre-trained checkpoints will be provided at [URL].