This paper introduces two complementary approaches to improve motif-scaffolding using an SE(3) flow matching model called FrameFlow. The first approach, motif amortization, trains FrameFlow with the motif as input using data augmentation to generalize to new motifs. The second approach, motif guidance, uses an estimate of the conditional score from FrameFlow to scaffold the protein without additional training. On a benchmark of 24 biologically meaningful motifs, the proposed method achieves 2.5 times more designable and unique motif-scaffolds compared to the state-of-the-art method RFdiffusion. The results show that FrameFlow with both motif amortization and guidance surpasses RFdiffusion in terms of structural diversity. The method is evaluated on a benchmark of 24 motifs, where FrameFlow-amortization achieves nearly double the number of unique scaffolds compared to RFdiffusion. The results also show that FrameFlow-amortization achieves higher diversity and faster inference than RFdiffusion and TDS. The paper also discusses the importance of measuring diversity alongside designability to detect mode collapse. The results demonstrate that FrameFlow-amortization achieves much more structural diversity than RFdiffusion. The paper concludes that the proposed methods are effective for motif-scaffolding and can be used with any flow-based model.This paper introduces two complementary approaches to improve motif-scaffolding using an SE(3) flow matching model called FrameFlow. The first approach, motif amortization, trains FrameFlow with the motif as input using data augmentation to generalize to new motifs. The second approach, motif guidance, uses an estimate of the conditional score from FrameFlow to scaffold the protein without additional training. On a benchmark of 24 biologically meaningful motifs, the proposed method achieves 2.5 times more designable and unique motif-scaffolds compared to the state-of-the-art method RFdiffusion. The results show that FrameFlow with both motif amortization and guidance surpasses RFdiffusion in terms of structural diversity. The method is evaluated on a benchmark of 24 motifs, where FrameFlow-amortization achieves nearly double the number of unique scaffolds compared to RFdiffusion. The results also show that FrameFlow-amortization achieves higher diversity and faster inference than RFdiffusion and TDS. The paper also discusses the importance of measuring diversity alongside designability to detect mode collapse. The results demonstrate that FrameFlow-amortization achieves much more structural diversity than RFdiffusion. The paper concludes that the proposed methods are effective for motif-scaffolding and can be used with any flow-based model.