生成式人工智能介入短视频创作的应用模式与效果研究——基于四个典型案例的分析

Research on the Application Modes and Effects of Generative AI in Short Video Creation—An Analysis Based on Four Typical Cases

摘要: 针对生成式人工智能(AI)深度介入短视频创作，但现有研究缺乏基于真实创作数据的效果验证这一理论缺口，研究旨在考察生成式AI在短视频创作中形成了哪些应用模式，以及这些模式对创作效率、传播效果与内容质量产生了何种实际影响。为此，以技术接受模型(TAM)为分析框架，引入技术社会形塑理论与创新扩散理论，采用多案例比较方法，遵循理论抽样原则，选取知识科普、AI短剧、品牌营销、影视解说四个典型案例。数据来源包括创作者工作日志与自述、蝉妈妈平台作品发布后30天传播数据、三位评估者对叙事完整性、视听表达和原创性的独立评分，以及各案例评论区前200条有效评论的人工编码。研究发现，AI介入短视频创作形成三种典型模式：单环节工具辅助型、人机协同增强型和全流程AI生成型。效率维度上，四案例制作周期缩短40%~83%，提升幅度与AI介入深度正相关；传播维度上，整体改善但模式分化，全流程型播放量增长逾90%，单环节型不足20%，AI短剧案例分享率由5.6%升至11.2%；质量维度上，呈结构性分化，叙事完整性和视听表达指标提升，原创性则在全部案例中下降。信任维度上，AI介入越深，来源质疑与真实性追问越突出，信任感知分化明显。据此，提出“效率–质量–信任”三维框架，揭示“效率提升–质量分化–信任重构–传播反馈”的循环机制，验证并拓展了TAM在短视频创作情境中的适用性，修正了AI普遍提升内容质量的既有预期，为创作者差异化选用AI、平台标识治理与监管分级提供了实证依据。

Abstract: Addressing the theoretical gap that generative artificial intelligence (AI) has deeply intervened in short video creation while existing research lacks effect validation based on real creative data, this study aims to examine what application modes generative AI has formed in short video creation and what actual impacts these modes have on creation efficiency, dissemination effects, and content quality. To this end, the study adopts the Technology Acceptance Model (TAM) as its analytical framework, introduces the Social Shaping of Technology (SST) theory and Diffusion of Innovations (DOI) theory, and employs a multiple-case comparative method. Following the principle of theoretical sampling, it selects four typical cases: knowledge popularization, AI short drama, brand marketing, and film/TV commentary. Data sources include creators’ work logs and self-reports, 30-day dissemination data of works on the Chanmama platform after publication, independent ratings by three evaluators on narrative completeness, audiovisual expression, and originality, as well as manual coding of the first 200 valid comments in each case’s comment section. The study finds that AI intervention in short video creation forms three typical modes: single-link tool-assisted mode, human-AI collaborative enhancement mode, and full-process AI generation mode. In terms of efficiency, the production cycles of the four cases are shortened by 40%~83%, and the degree of improvement is positively correlated with the depth of AI intervention. In terms of dissemination, overall improvement is observed, but patterns diverge: the full-process mode sees a playback growth of over 90%, while the single-link mode sees growth of less than 20%; the share rate of the AI short drama case rises from 5.6% to 11.2%. In terms of quality, structural differentiation appears: narrative completeness and audiovisual expression indicators improve, while originality declines in all cases. In terms of trust, the deeper the AI intervention, the more prominent the questioning of sources and authenticity, and trust perception shows clear differentiation. Accordingly, this study proposes an “Efficiency-Quality-Trust” three-dimensional framework, revealing the cyclical mechanism of “efficiency improvement - quality differentiation - trust reconstruction - dissemination feedback”. It validates and extends the applicability of TAM in the context of short video creation, corrects the existing expectation that AI universally improves content quality, and provides empirical evidence for creators’ differentiated selection of AI, platform labeling governance, and regulatory classification.