Processes and Adaptation of AI Tools for Novice Content Creators

1. Target contents

To support novice content creators aiming to improve the efficiency and quality of their video production, the practical application of AI tools is essential. According to prior research [2], content creators can be categorized into three types: participatory entertainers (Type 1), knowledge sharers (Type 2), and highlight capturers (Type 3). Type 1 creators draw inspiration primarily from everyday life and often feature themselves as the main subjects in their videos—for example, through dance challenges or role-playing scenarios [3]. Type 3 creators, on the other hand, reprocess and edit previously published video materials, focusing on extracting and compiling highlights from sources such as television shows or films [3].

Both Type 1 and Type 3 rely heavily on video footage—either newly recorded or pre-existing—and involve identifying key moments to highlight in the final product. As such, their production workflows are similar. With the aid of automated video editing tools powered by AI, the task of identifying highlight moments can be significantly streamlined, requiring minimal human intervention and thereby reducing production time considerably [10].

In contrast, Type 2 content does not typically rely on existing videos. Instead, it emphasizes the delivery of new and informative content to viewers, such as scientific explanations, educational tutorials, or commentary [3]. For this type, scripting and storyboarding are crucial to clearly communicate the intended message. Traditionally, creating scripts and storyboards has been a time-consuming process. However, the integration of AI tools—such as generative text models like ChatGPT—has simplified these tasks. AI can rapidly generate structured content, including full video outlines and scripts. Moreover, AI-driven translation tools now enable creators to easily localize their content into multiple languages, offering a level of convenience and scalability that surpasses traditional production workflows.

2. Co-Creation Process with Generative AI tools for Content Creation

According to collaborative creativity theory, collaboration enhances creativity because innovation is not the result of a single flash of insight but rather emerges through a series of interconnected ideas. Real innovation, therefore, is typically the product of collective input rather than individual effort [12]. In this context, AI can be conceptualized as an additional cognitive agent—or “another brain”—that supports and amplifies human creativity. Based on the distinct characteristics of various content types, we implement tailored Human–AI co-creation workflows that strategically leverage AI tools. The overall process comprises two primary stages: AI-assisted generation and manual refinement.

In the AI generation phase, multimodal assets—such as images, video clips, and audio tracks—are produced using generative AI tools. Users are then prompted to evaluate the intermediate outputs and proceed with manual editing to finalize the content. This co-creative workflow is adapted to suit the three different content types.

Type 1 content begins with original footage recorded by the creator, while Type 3 content utilizes pre-existing, publicly available videos. Both types involve identifying and extracting relevant segments from raw video material, and therefore follow a similar initial workflow. Using AI-powered auto-editing tools, the system automatically detects highlight moments and compiles them into a coherent video composition. This automation substantially reduces production time. If the initial output is unsatisfactory, users can iterate by reapplying the auto-editing process. Once highlight clips are finalized, the process moves into post-production, which involves manual editing and rendering using standard video editing software (see Fig. 1).

Fig. 1.

Co-Creation Process with Generative AIs

Type 2 content follows a more generative process and typically begins from scratch. It starts with the generation of scripts, storyboards, song lyrics, and prompts using a language model such as ChatGPT. These prompts are then used in AI image generators to create visual assets, which are subsequently input into a video generation tool to produce short clips. Simultaneously, a music generator creates melodies based on the generated lyrics. Once all assets—images, videos, and audio—are created, users assess their quality. If any component is deemed unsatisfactory (e.g., irrelevant scenes or mismatched audio), it is refined by returning to the corresponding generation stage. When all components meet the desired quality standards, they are assembled in video editing software to produce the final product, ready for distribution (see Fig. 1).

1. Analysis approach

To better understand individual experiences with the proposed content creation process, we adopted a triangulation framework for analysis [11]. This methodological approach allows for a comprehensive evaluation of the effectiveness of the process by comparing multiple data sources. Specifically, the study employed three forms of data for triangulation: Questionnaires, which captured participants’ self-rated assessments of content quality, perceived efficiency, and tool usability; Interviews, conducted before and after the workshop, which revealed user motivations, perceived challenges, and changes in perception; Production outputs, which served as tangible evidence of the content produced using AI tools and were evaluated for quality and consistency. This triangulated approach enabled cross-validation of findings and facilitated deeper insights into participants’ creative experiences and process-related challenges.

To further understand participants’ creative processes, needs, and obstacles, semi-structured interviews were conducted both before and after the online workshop. The pre-workshop interviews ensured that participants’ questions could be addressed during the workshop and allowed us to verify whether the proposed co-creation process supported their content production effectively. During the pre-workshop interviews, we introduced the research focus and collected background information, including participants’ previous video production experiences, production motivations, tools previously used, desired outcomes, and the social media platforms they intended to use.

This small-scale workshop was designed as a pilot study to evaluate the effectiveness of the tailored Human–AI co-creation process. Participants were recruited based on the following criteria: (1) individuals planning to launch a new channel on platforms such as YouTube, Instagram, or TikTok; (2) individuals with limited prior video production experience; and (3) individuals currently residing in South Korea. Recruitment was conducted via LINE messenger groups, specifically through a post in the “Taiwanese in Korea” community. Four individuals (three female, one male) met the inclusion criteria. The average age of participants was 32 years, and their prior production experience ranged from six months to three years, with a mean of 1.6 years (see Table 1).

Table 1.

Interview before the workshop

Semi-structured interviews were conducted remotely via Zoom prior to the workshop. Each session lasted approximately 30 minutes and was organized into three sections. The first section introduced the study’s objectives and scope. The second section explored participants’ video production backgrounds, including the tools they had used, their motivations, challenges faced, and current needs. The final section addressed participants’ expectations regarding the content creation process and the intended outcomes. All interviews were conducted in Taiwanese Mandarin with participants’ consent, and each participant planned to operate their channel primarily in Taiwanese Mandarin, with some also intending to provide bilingual subtitles. Participants were compensated 50,000 KRW for each interview session.

Subsequently, the workshop was conducted remotely via Zoom video conferencing and lasted approximately two hours. As a reference, a sample channel introduction video was created for participants to review. This sample video included AI-generated elements such as background music, narration, bilingual subtitles, images, and short video clips—all produced using the AI tools introduced during the workshop. The workshop was organized into the following sections: 1) Generating song lyrics, video scripts, storyboards, and AI prompts using ChatGPT; 2) Generating images; 3) Creating short video clips; 4) Combining these short clips into a longer video using video editing software; 5) Rendering the final video for publication; and 6) Providing supplementary explanations tailored to participants’ individual learning needs, including demonstrations of additional AI tools.

Each step was accompanied by a brief tutorial before participants began working on their own projects. After the workshop, participants submitted their completed works exclusively for research purposes. The entire workshop was conducted in Taiwanese Mandarin and was recorded with participants’ consent.

Finally, follow-up interviews were conducted to assess whether the process was effective during participants’ content production. All participants reported that they were able to quickly create their own videos by following the proposed workflow.

2. Analysis Result

Through the workshop, we observed that participants had diverse objectives in using generative AI tools for content creation, and their workflows reflected the corresponding content types. Participant 1 (P1), who aimed to produce travel and lifestyle videos, typically required original footage. During the workshop, P1 followed the Type 2 process to learn how to generate videos from scratch and incorporated AI-generated narration in the final output. Participant 2 (P2) focused on translating existing videos. By following the Type 3 process, P2 learned to use AI tools to automatically generate bilingual subtitles within seconds. Participant 3 (P3), who intended to create cooking tutorial videos, also followed the Type 3 process and used AI to efficiently extract highlight moments from existing video footage. Participant 4 (P4) wanted to produce pet-related content. Following the Type 1 process, P4 learned to use AI tools to identify highlight moments in existing pet videos. Additionally, P4 experimented with the Type 2 process to generate new pet-related content using AI-based media generation tools.

Post-workshop interviews revealed a range of perspectives on the content creation process. All participants provided generally positive feedback, particularly noting the usability of the tools and the perceived increase in production efficiency. They also expressed satisfaction with the breadth of content covered during the workshop. Two participants specifically appreciated learning about multiple tools in a short period.

However, several limitations were also identified. P1 noted that the AI-generated narration did not sound natural or reflect the accent of native Taiwanese speakers. P2 highlighted that the AI-generated images failed to accurately distinguish between Korean and Taiwanese cultural elements. As a result, the visuals lacked identifiable cultural characteristics, and some included fictitious or nonsensical text. This points to an ongoing challenge in culturally nuanced AI content generation. P3 expressed a preference for the AI music generation tool Suno, although they noted issues with visual consistency—such as the main character’s appearance changing across scenes and instances of anatomical inaccuracies, including six fingers. Interestingly, P3 also remarked that due to the apparent convenience of the tools introduced, they were considering investing more in AI tools in the future. P4 encountered challenges related to script refinement and storyboard adjustments. After multiple iterations with ChatGPT, P4 was eventually able to achieve the desired results. P4 also commented that the AI tools offered a wide range of creative directions, which helped inspire and enhance human creativity.

We also found that participants spent less time creating content with the assistance of generative AI, although the quality of the resulting outputs varied considerably. Participants who marked “M” on their scripts indicated that they needed to modify the content after it was initially generated by ChatGPT. The quality of the generated images and videos was rated across a broad spectrum, reflecting the variability in AI output. For example, Participant 1 (P1) reported satisfactory results from text-to-image generation but was dissatisfied with the quality of the AI-generated video. Conversely, Participant 2 (P2) experienced the opposite: although some of the generated images were unsatisfactory, the resulting video met expectations. The “None” label in the auto-editing section indicated that the participant did not utilize the automated editing feature for their video. Despite differences in content quality, all participants reported that the use of AI tools resulted in a time savings of up to 50% compared to traditional production workflows. Since all the AI tools introduced during the workshop offered free trial versions, participants were able to experiment with multiple platforms and decide which tools to adopt based on the quality of the outputs and their individual needs.

Table 2.

Interview after the workshop

Table 3.

Total time consuming & Self-rating Quality Chart