Boost the Identity-Preserving Embedding for Consistent Visual Generation

Text-to-image models have advanced high-fidelity content generation, but their inability to maintain subject consistency hampers realistic applications. Existing training-based methods rely on heavy computation and large datasets; while training-free approaches demand excessive memory or complex auxiliary modules. In this paper, we first reveal a key property overlooked in prior works that the identity-relevant signals, termed Identity-Preserving Embeddings (IPemb), are implicitly encoded in textual embeddings of frame prompts. To address the consistent T2I generation with the IPemb embedding, we propose Boost Identity-Preserving Embedding (BIPE), a training-free yet plug-and-play framework that explicitly extracts and enhances the IPemb. Its core innovations are two complementary techniques: First, Adaptive Singular-Value Rescaling (adaSVR) applies singular-value decomposition to the joint embedding matrix of all frame prompts, amplifying identity-centric components while suppressing frame-specific noise. Second, Union Key (UniK) further reinforces consistency by aligning the T2I backbone’s image-text attention across the entire generation sequence. Experiments on the ConsiStory+ benchmark demonstrate BIPE outperforms existing methods in both qualitative and quantitative metrics. To address the gap in evaluating a broader range of scenarios with diversified prompt templates, we introduce a DiverStory benchmark to further confirm our scalability.

AI image generation models can already produce high-quality images, but they often struggle to keep the same character consistent when generating a sequence of story images. For example, the same person may no longer resemble themselves across different scenes. Existing methods typically require additional training, large amounts of data, or substantial GPU memory, which limits their practicality. We propose BIPE, a training-free method that can be directly integrated into existing text-to-image models. BIPE is based on the observation that the text prompts for different images in a sequence already contain shared information that preserves character identity. It extracts and enhances this information, enabling the model to focus more consistently on the same subject when generating multiple images and thereby reducing variations in character appearance. BIPE helps users generate character-consistent story images, animation storyboards, and long-form visual content more easily. Experiments show that it preserves subject identity effectively while maintaining good adherence to the text description of each image, and it also generalizes to more natural and diverse prompt formulations.