On March 28th, the Lingnan Digital Creative Building Construction Project for High-Quality Development Conference was convened at the Yangcheng Creative Industrial Park. The event witnessed the official inauguration of the Lingnan Cultural Big Data Center and the debut of the Lingnan Cultural Large Model.
The center will operate under a structural framework termed "onefoundational infrastructure, twoplatforms, three strategic levers, and X application scenarios". By consolidating cultural lineage, intangible cultural heritage (ICH), folk traditions, and regional customs, it will construct three core components: the Lingnan Cultural Large Model, an online data service platform, and an offline digital creativity center. These initiatives will drive the development of over 100 intelligent applications, such as Digital Village History Museums, aI-Powered Cultural Tourism Self-Guided System, and Cross-Border Intellectual Property (IP) Trading Platforms, thereby accelerating the digitization and revitalization of cultural heritage.
For offline engagement, the 120,000-square-meter Lingnan Digital Creative Center will employ VR/AR technologies and spatial computing to craft an immersive Lingnan Digital Culture Experience Hall. On the digital front, the Lingnan Culture Data Service Platform, equipped with "Data Supermarket" and "Co-Creation Community" modules, will highlight Lingnan's cultural uniqueness, interface with the National Cultural Network, facilitate bidirectional data exchange with the National Database, and pioneer China's first Cross-Border Cultural Data Rights Authentication and Trading Mechanism.
As a flagship cultural publication in the Lingnan area, Yangcheng Evening News has pursued "AI+" media innovations. Notable collaborations include partnering with NetEase Games AI Lab to introduce Guangdong's first AI humanoid news anchor, co-establishing the AI Converged Media Laboratory with Tencent, and embedding DeepSeek into its Young Pai App.
Now, the newspaper has unveiled a landmark "AI + Culture" initiative, the Lingnan Cultural Large Model, jointly developed by Yangcheng Evening News and the Zhimei Yuntu Joint Laboratory. This model represents China's first multimodal reasoning large model dedicated to cultural verticals.
This large model has surpassed OpenAI in cold-start multimodal evaluations (covering image and video modalities), demonstrating domain-specific expertise and enhanced knowledge comprehension accuracy in Lingnan culture. It establishes a new technical paradigm for vertical industry-oriented multimodal reasoning large model in China."
According to technical disclosures, the model's training corpus integrates high-dimensional heterogeneous data sources, including authoritative Lingnan Culture Encyclopedias, reports from Yangcheng Evening News's cultural lineage series, And extensive unstructured audiovisual archives. These resources collectively construct a knowledge graph encompassing multidimensional regional art forms such as Cantonese opera, lion dance, paper-cutting, pottery, And embroidery as well as historical architectural data. Deng Libang, director of the Lingnan Cultural Large Model Laboratory, revealed that the multimodal content spans 12 major categories and over 200 subcategories.
For example, when users inquire about Guangzhou Xiguan's traditional Tanglong, A traditional sliding wooden gate, the model comprehends the cultural context and delivers precise explanations. In another scenario, if a user uploads an image of Lingnan cuisine and asks, "Which dishes are salty?", the model not only identifies the depicted food items but also infers cooking methods to provide validated answers.
Imagine learning Cantonese or crafting ICH items like paper-cutting and shadow puppetry with the Lingnan Cultural Large Model, entering a story concept to auto-generate narratives, And then transforming them into Cantonese opera-style theatrical scripts. It is estimated that between 2026 and 2028, the model will achieve full multimodal interaction and content generation, empowering over 5,000 micro and small-sized cultural enterprises to drive innovation.
3月28日，岭南数字创意大厦新基建高质量发展大会在广州羊城创意产业园举行，现场正式挂牌成立岭南文化大数据中心，并发布了岭南文化大模型。
岭南文化大数据中心将以“1底座、2平台、3抓手、X场景”为架构，通过整合文脉、非遗、历史、民俗等文化资源，建设岭南文化大模型、线上数据服务平台和线下数字创意中心，打造数字村史馆、AI文旅自助导览、跨境IP交易所等百余项智能体应用，推动文化遗产数字化保护与活化利用。
在线下端，12万平方米的岭南数字创意中心将融合VR/AR、空间计算技术，打造出沉浸式的岭南数字文化体验馆。在线上端，上架“数据超市”“共创社区”等功能的岭南文化数据服务平台，也将聚焦岭南文化特色，并接入国家文化专网，实现岭南文化数据与国家总库双向流通，并建立全国首个跨境文化数据确权交易机制。
作为岭南文化大报，羊城晚报的“AI+”智媒尝试从未停步。携手网易互娱AI Lab推出广东首个媒体真人AI主播，和腾讯共创AI融媒体实验室，羊城派客户端接入DeepSeek……
站在新的风口上，羊城晚报交出了一份“AI+文化”的跨界答卷——由羊城晚报和智媒云图联合实验室研发的岭南文化大模型，这也是国内首个文化领域的多模态推理大模型。
这一大模型在图片、视频的多模态冷启动评测中领先OpenAI，在岭南文化这一垂直领域内拥有更为专业及准确的知识理解能力，开创国内垂直行业多模态推理大模型的技术新范式。
据介绍，该模型训练语料库基于高维度异构数据源整合，包括权威的岭南文化辞典、《羊城晚报》文脉专题报道以及大量非结构化的音视频影像资料，覆盖了粤剧、舞狮、剪纸、陶艺、刺绣等多维度地方艺术形态与历史建筑信息的知识图谱。岭南文化大模型实验室主任邓立邦给出一组数据：大模型的数据有12大类、200多子类的多模态内容。
当用户问起广州西关的木门“趟栊”，大模型能理解，并给出准确的文化解释；当用户上传一张岭南美食图片，并且问“哪些食物是咸的”，模型不仅能识别图片中的食物，还可以思考食物的具体做法，进行推理给出正确答案……
想象一下，跟着岭南文化大模型一起学习粤语，学习剪纸、皮影等非遗制作；输入一个故事的主题，用大模型生成一段你想要的故事，再用粤剧的艺术形式生成一段戏曲……据估算，岭南文化大模型在2026年至2028期间将实现多模态交互及生成，将支撑超5000家小微文化企业创新发展。