Timeline:
I designed based on the data and insights gained from operating the AI Voiceover tool ‘Genny’ and by researching various presentation tools familiar to the target users.
Insight #1
Designing for Attention
As employees experience fatigue from reading long blocks of text, visual and interactive content have become essential for effective communication.
Insight #2
Efficiency Starts with Script
Storyboard and script creation takes up 50% of the total work time when converting long text into a video presentation.
Insight #3
The Cost of Switching
There is no tool that handles all stages of video content creation. Going back and forth between multiple tools not only decreases worker efficiency but also increases costs.
Key #1
Perfect Story Structure Through AI Document Analysis
Key #2
Customizable and Visually Appealing
Slide Templates
Key #3
All-in-one, Transforming a Complex Process into a Seamless Workflow
① Research
Since there are few Document-to-Video Presentation tools, I focused on designing an easy-to-use tool with a low learning curve.
After conducting user interviews, I found that the target users are familiar with PPT tools. Given that the Doc2Vid feature was new to them, I implemented a UI inspired by PPT tools to create an intuitive interface that enhances the user experience and removes any barriers users might face when encountering new technology.
② Key UX Strategies
Outline
Through extensive research and testing, my team and I identified the limitations of the AI technology. I realized that solving these limitations through design was key to the entire process.
To solve this, I came up with the 'Outline' stage - offering users a draft of the presentation structure before the final project is generated. This step not only ensures alignment with the imported document and the user’s intent, but also helps prevent potential AI errors early in the process.
Scene Based Editing
Since the target users are used to presentation tools, I applied the concept of slides into 'Scene-based editing.' While assets within each scene can be finely adjusted using a timeline and preview, the scenes are displayed as a list on the left - just like slides.
This approach allows non-professional video creators to intuitively understand the tool, making it easy to pick up and use without extra training.
③ Templates
Doc2Vid relies heavily on AI technology, making template creation was a crucial task. I repeatedly tested and adjusted pre-designed templates to fit the LLM format, ensuring users could easily achieve the final output with minimal effort.
Additionally, I identified limitations in Genny’s legacy, particularly with the timeline and preview features, which could potentially degrade the user experience when using a premade template. To address these technical constraints, I reorganized the asset layers within the templates from a UX perspective, enhancing both functionality and the overall user experience.
Feature #1
Doc 2 Vid Step 1: Document Import
Feature #2
Doc 2 Vid Step 2: Analyzed Document
Feature #3
Doc 2 Vid Step 3: Auto Outlined
Feature #4
Doc 2 Vid Step 4: Template Gallery
Feature #5
Scene Based Editing Project
While working on AI features like Auto Subtitles, AI Generator, and AI Voice for Genny, I gained a lot of insights into effective communication with the development team, as well as essential considerations when working with AI Open Source and APIs. This experience allowed me to quickly start Keytake project. However, using LLM and templates, and lack of LLM testing variables, was quite challenging.
The LLM itself was smart, but creating a product based on LLM required a lot of careful thought and consideration. Honestly, it was really difficult. Despite both projects being LLM-based, this felt completely different from planning the AI Writer. Even for minor details, like importing a page via URL and converting it to HTML, I constantly had to think about how to handle Alt text and other elements to achieve better output. The most challenging part was positioning the LLM data within a fixed template format. Even just before the product launch, I was still testing and revising the template design, putting all my energy into solving LLM formatting issues.
However, the time spent considering new flows based on LLM test results and exploring new error cases was rewarding. It was interesting to think about how to make the limitations of AI functionality more user-friendly. (Was it really fun? Maybe… but I can confidently say that I learned something new!)
