KuaiKan Comics' Efforts in AI Fine-Tuning Open-Source Model for 2D Vertical Model

TapTechNews on July 13th reported that Tian Guicheng, a senior algorithm engineer of the KuaiKan Search R & D team, disclosed at the Zilliz UserTech event on July 11th that KuaiKan Comics is currently attempting to fine-tune based on an open-source large model to train a vertical large model in the field of two-dimensional (2D) .

TapTechNews learned from KuaiKan that KuaiKan claims that it has a very rich RAG (Retrieval-AugmentedGeneration) application scenario internally. Currently, it has made substantial progress in AI intelligent Q&A and IP character interaction two scenarios, and the relevant AI products are currently in internal testing, and users will be able to experience the functions of the new products soon in the future.

According to the introduction, the implementation plan behind IP character interaction is a role-playing large model + knowledge base RAG, and the knowledge base is divided into original plot knowledge base and user personalized memory knowledge base. The official also said, We hope to derive IP character interaction applications through these IPs. Users can have immersive chat with the characters in the IP, not a pure virtual character chat, which is different from the common virtual role-playing applications in the market for KuaiKan IP character interaction.