3 papers
cs.CV2026
SafeStyle: Calibrated Style Residual Injection for Controllable Style-Leakage Trade-off in Diffusion Stylization
Zhangping Yang, Min Li, Song Yan +4
Reference-guided diffusion stylization aims to transfer visual style from a reference image while preserving the semantics specified by a text prompt. However, image conditioning o…
cs.CV2026
Breaking the Synthetic-Real Domain Shortcut for Training-Free Generative Replay-based Class Incremental Learning
Tao Zhang, Qixuan Fan, Yiyuan Liang +7
Class-incremental learning (CIL) requires models to continuously acquire new knowledge while avoiding catastrophic forgetting. While exemplar replay is effective, it raises concern…
cs.CV2025
Determinism of Randomness: Prompt-Residual Seed Shaping for Diffusion Generation
Song Yan, Wei Zhai, Chenfeng Wang +9
Diffusion models start generation from an isotropic Gaussian latent, yet changing only the random seed can lead to large differences in prompt faithfulness, composition, and visual…