Comic Fast View: Training a Vertical Large Model in the ACG Domain for Immersive Chatting with IPs In the realm of anime, comics, and games (ACG), the rapid development of technology has led to the creation of large-scale vertical models. These models are specifically designed to provide users with an immersive experience when interacting with their favorite intellectual properties (IPs). The ability to engage in natural and lifelike conversations with characters from these IPs enhances the overall enjoyment and engagement of the fans. This technological advancement is a testament to the continuous innovation and progress in the ACG industry.

On July 11th, Kuaikan Comics announced that they are exploring fine-tuning of open-source large models to train a vertical large model for the two-dimensional (2D) domain.

Kuaikan Comics utilizes large language models (LLM) and retrieval-augmented generation (RAG) technology to improve the conversion rate of work searches and inject new vitality into the search process. They have adopted a "fine-tuned large model + RAG enhancement" strategy, building internal PGC (Professional Generated Content) specialized knowledge bases, UGC (User-Generated Content) post knowledge bases, and Q&A knowledge bases. By intent recognition, they determine Query routing, reduce generation time, and increase response speed for answers. During the recall and reranking stages, Kuaikan applies fine-tuned Embedding models and Rerank reranking models, improving recall rates and ranking indicators.

As the largest comic platform in China, Kuaikan Comics boasts over 13,000 comic works and 80% of top-quality IP. In terms of IP character interaction, Kuaikan Comics aims for immersive chatting experiences with characters within IPs, differing from common virtual role-playing applications on the market. The underlying implementation for IP character interaction is a role-playing large model + knowledge base RAG, including original plot knowledge bases and user personalized memory knowledge bases. The original plot knowledge base is an advantage of Kuaikan.

Since its inception, Kuaikan Comics has been focused on the development of AI technology, driving the application of new technologies through research and exploration to empower the innovative development of the comic industry. In 2014, they pushed for the popularity of strip comics, in 2019 they collaborated with universities to explore the application of AI in comic creation and reading, and in April 2023, they established the AIGC (AI-assisted Generation of Content) division. In July, they announced the start of a new technology phase, launching an AI digital creation workspace to help creators find inspiration, assist with screenwriting, and auto-coloring, enhancing user experience and expanding content production capacity.

Kuaikan Comics believes that the implementation of AIGC will enhance the competitiveness of comics, expand user experience and production capacity through new technologies, and create new commercial value for the industry. With the development of generative AI technology, artificial intelligence is bringing about a revolution in content creation, improving the efficiency of content production, reducing costs, and automatically generating personalized content based on user needs and preferences, promoting the expansion of innovative applications.