EditID: Training-Free Editable ID Customization for Text-to-Image Generation

TL;DR

EditID tackles the challenge of editable ID customization in text-to-image generation by introducing a training-free paradigm within the DiT architecture. It decouples ID features into mapping (local) and shift (regional) features, and couples them with a dynamic ID integration mechanism to achieve prompt-driven edits while preserving ID fidelity. The method shows state-of-the-art editability on long prompts and introduces IBench, a configurable evaluation framework that jointly assesses T2I quality, ID consistency, and editability. Practically, EditID enables flexible character edits (pose, hairstyle, age, etc.) without retraining, which is valuable for story generation, character design, and other creative workflows.

Abstract

We propose EditID, a training-free approach based on the DiT architecture, which achieves highly editable customized IDs for text to image generation. Existing text-to-image models for customized IDs typically focus more on ID consistency while neglecting editability. It is challenging to alter facial orientation, character attributes, and other features through prompts. EditID addresses this by deconstructing the text-to-image model for customized IDs into an image generation branch and a character feature branch. The character feature branch is further decoupled into three modules: feature extraction, feature fusion, and feature integration. By introducing a combination of mapping features and shift features, along with controlling the intensity of ID feature integration, EditID achieves semantic compression of local features across network depths, forming an editable feature space. This enables the successful generation of high-quality images with editable IDs while maintaining ID consistency, achieving excellent results in the IBench evaluation, which is an editability evaluation framework for the field of customized ID text-to-image generation that quantitatively demonstrates the superior performance of EditID. EditID is the first text-to-image solution to propose customizable ID editability on the DiT architecture, meeting the demands of long prompts and high quality image generation.

Figures (15)

• Figure 1: We introduce EditID, a training-free ID customization approach. EditID achieves better editability compared to similar methods. It demonstrates excellent editability in long prompts (where action prompts are marked in red) and aligns well with Flux's T2I.
• Figure 2: Overview of the EditID Framework. The upper half of the framework depicts the DiT-based image generation process. The lower half represents the character feature branch, which is divided into three parts. The first part is the ID feature extraction Module, responsible for extracting global and local features, generating mapping features. The second part is the ID feature fusion module, tasked with fusing the mapping features, producing shift features at this stage. The third part is the ID feature integration module, which implements the dynamic ID embedding mechanism design.
• Figure 3: Detailed diagram of the character feature extraction module. The blue facial features and the dark green CLS token features from EvaCLIP together form the global features. The yellow features represent the mapping features, with layers 4, 8, 12, 16, and 20 in EvaCLIP being the original mapping features. Gray indicates the unselected features.
• Figure 4: Impact of global and local features on generation editability. Gray indicates unselected features, set to zero.
• Figure 5: Impact of global and local features on generation editability. Gray indicates unselected features, set to zero.