ArtiFade: Learning to Generate High-quality Subject from Blemished Images

ArtiFade (Ours) Textual Inversion Input images ArtiFade (Ours) DreamBooth Input images Visible artifacts Invisible artifacts Figure 1. Blemished subject-driven generation with our ArtiFade and vanilla subject-driven methods. We display images generated using ArtiFade and Textual Inversion [15] on watermark artifacts on the left, and ArtiFade and DreamBooth [44] on adversarial noise artifacts [48] on the right. In contrast to the poor performance of Textual Inversion and DreamBooth, which are negatively affected by the visible or invisible artifacts, ArtiFade produces much better fidelity of the subject with high-quality generation.