Entity summarization in knowledge graphs (KGs) aims to generate concise summaries of entities by selecting the most relevant facts (triples) from structured KG data, such as DBpedia or Wikidata. Existing methods focus on selecting triples that are directly present in KGs, which inherently contain incomplete data. To address this limitation, we propose Ants, an abstractive approach that generates entity summaries beyond KG triples. Our approach first identifies relevant entity-related triples using KG embeddings. Furthermore, since KG embeddings struggle with literal values (e.g., numbers, dates), we integrate a large language model (LLM) to augment plausible triples with literal objects. For evaluation, we used Essum, a silver-standard dataset combining ESBM-DBpedia and FACES benchmarks. Experimental results demonstrate that Ants outperforms baseline models, achieving a BLEU score of 5.78, BERTScore F1 of 0.84, and MoverScore of 0.54.