Abstract: The current state-of-the-art text-to ... TextReg regulates the fine-tuning of cross-attention modules with simple text prompts without identifiers, which avoids the usage of real images as ...