PromptTCP: From prompting to knowledge-driven concept orchestration for co-creative traditional Chinese painting

In professional artistic creation with Generative AI, creators struggle to translate their multi-dimensional, implicit artistic intentions into the linear instructions required by generative models, creating a significant semantic gap. This challenge is particularly acute in traditional Chinese figure painting, which emphasizes sophisticated brushwork, stylistic lineage, and cultural conception. To address this, we present PromptTCP, a domain knowledge-driven human-AI co-creation system. By integrating a domain-specific knowledge graph (TOBEST-SemKG) encoding six artistic dimensions with a visual“Semantic Canvas” for hierarchical concept organization, PromptTCP transforms prompt engineering from a high-load linguistic task into a co-creative process of conceptual orchestration and visual feedback. A mixed-methods study with 12 participants demonstrates that PromptTCP significantly reduces user frustration and effort while enhancing satisfaction. Furthermore, the generated prompts and images surpass a text-based baseline in richness, cultural relevance, and artistic conception. Our findings provide empirical support for culture-specific AIGC applications and offer a transferable methodology for designing co-creative systems for professional domains.