The paper introduces AMP-Diffusion, a novel latent space diffusion model designed for generating antimicrobial peptides (AMPs). This model leverages the capabilities of the state-of-the-art protein language model (pLM), ESM-2, to de novo create functional AMPs. The study evaluates AMP-Diffusion against experimentally validated AMPs, demonstrating that the generated peptides align closely in pseudo-perplexity and amino acid diversity, and exhibit physicochemical properties similar to natural sequences. The framework's effectiveness is further validated through external classifier assessments and comparisons with other models like HydrAMP, PepCVAE, and AMPGAN. The results highlight the biological plausibility of the generated sequences and suggest potential applications in protein design and therapeutic interventions. The paper also discusses the integration of diffusion models with pLMs, emphasizing the need for future research in this area.The paper introduces AMP-Diffusion, a novel latent space diffusion model designed for generating antimicrobial peptides (AMPs). This model leverages the capabilities of the state-of-the-art protein language model (pLM), ESM-2, to de novo create functional AMPs. The study evaluates AMP-Diffusion against experimentally validated AMPs, demonstrating that the generated peptides align closely in pseudo-perplexity and amino acid diversity, and exhibit physicochemical properties similar to natural sequences. The framework's effectiveness is further validated through external classifier assessments and comparisons with other models like HydrAMP, PepCVAE, and AMPGAN. The results highlight the biological plausibility of the generated sequences and suggest potential applications in protein design and therapeutic interventions. The paper also discusses the integration of diffusion models with pLMs, emphasizing the need for future research in this area.