#Japan #AI tools #Tokyo #PRIZMA #Manga Creation

Unveiling the Challenges of AI-Generated Content

AI technology has become an integral part of many businesses, aimed at improving efficiency and streamlining workflows. However, a recent survey conducted by PRIZMA has shed light on an uncomfortable reality: many users are facing significant challenges regarding the consistency and accuracy of AI-generated outputs.

According to the survey, which involved participants aged 20 to 50 who regularly utilize AI for producing images or videos, over 90% are required to make multiple revisions on their projects. This has resulted in time losses ranging from several hours to a staggering level, predominantly in fields like image generation and manga creation. The findings raise important questions about the implications of relying on AI in professional settings.

The Burden of Revisions

When asked about the time and effort spent on AI-generated content revisions, more than 70% of respondents reported experiencing substantial burdens. Specifically, 20.1% stated they felt "very much" burdened by the revision work, while 56.3% felt a "slight burden." This indicates a concerning shift where a tool intended for efficiency appears to be creating more workload instead.

Users expressed frustrations regarding the inability to make specific amendments without undesired alterations to the entire output. Various issues arose during the revision process, such as the following:

• - Output Discrepancies: 44.2% of users reported frustration when trying to correct minor imperfections, only to find that entire sections were altered or lost.
• - Directional Misalignment: 41.5% experienced difficulties in maintaining the original intent after multiple iterations.
• - Misinformation: 26.3% encountered the challenge of "hallucinations," wherein incorrect information appeared in the generated content, necessitating further verification efforts.

These complications illustrate a clear disconnect in users’ expectations of AI and the operational realities they encounter. This unfulfilled promise of seamless content generation is contributing to overall dissatisfaction with AI tools.

Frequency of Revisions

The survey results also highlighted the average number of edits required for AI-generated outputs:

• - More than half of the respondents (60%) reported needing to revise their work four or more times, with some going as far as giving up on AI corrections altogether in favor of manually recreating their work. Only 3.1% stated that they achieved their desired output in one try.
• - Notably, a significant number of users (23.2%) mentioned that they require between 2 to 3 revisions on average, indicating that frequent alterations are now a normative experience amid AI’s supposed benefits.

Time Lost to Revisions

The impact of these revisions on productivity was also explored. When questioned about the time lost per project due to AI-induced revisions, the results were quite telling:

• - The majority (27.8%) felt that their time loss per project was between 1 to 3 hours, with 22.0% estimating between 30 minutes to 1 hour, and 14.5% suggesting a loss of 10 to 30 minutes.

This data exemplifies how something designed to save time has instead become a source of inefficiency, undermining the original objective of implementing AI.

Image Generation vs. Text Generation Challenges

When distinguishing between the ease of revising textual content versus visual content, around 69.3% of users indicated that image generation posed a greater struggle. Creating visual representations often entails starting from scratch due to the involvement of complex prompts, resulting in unwanted changes in the overall composition and background. Conversely, textual adjustments can often be made with simple edits, showcasing a disparity in user experience across different types of content creation.

Tools Being Used

In exploring what tools users employ for image generation, the survey revealed that:

• - ChatGPT leads with 66.4% of usage.
• - Other notable mentions include Gemini/Nano Banana (40.8%) and Microsoft Designer (18.0%).

This indicates a trend where users lean towards familiar multi-functional AI tools, even for tasks as specialized as image generation.

Manga Creation Insights

An intriguing aspect of the survey was the interest in manga creation, specifically its challenges when utilizing AI. Participants who had attempted to generate manga content reported several issues:

• - Creating characters with consistent features, such as specific poses or expressions, was seen as a significant hurdle (50%).
• - Maintaining accuracy regarding important elements, like brand logos within commercial contexts, adds another layer of complication (44.4%).

The findings suggest that younger generations are particularly eager to experiment with AI for this type of creative storytelling. However, without essential tools that permit intuitive adjustments and controls, the effectiveness of AI in manga production remains limited.

Conclusion

This survey illuminates the struggles prevalent in the current landscape of AI-generated content, where the expectations do not align with real-world experiences. As professionals adapt to AI tools, addressing the difficulties of revision processes—especially in image and manga generation—stands as a critical need for enhancing user satisfaction. For AI technology to achieve its intended purpose of streamlining workflows in business, developers must focus on refining the post-generation editing capabilities that users seek, ultimately ensuring smoother and more effective outcomes.