You can also find this content on ESTsoft’s official LinkedIn and Facebook.
In Part 1 [Itself] STUDIO PERSO ①: Delivering a consistent message with AI avatar videos made easily we looked at how to create professional AI human videos without filming the features of STUDIO PERSO and the stories of the developers behind it.
In Part 2, we meet the people behind the service. How does the STUDIO PERSO team work as they rapidly experiment, use the product themselves, and continually improve it? We asked Sangjin and Heeju about daily life at STUDIO PERSO and what lies ahead.
Q. Everyone on the STUDIO PERSO team is a developer. What sets the way your team works apart?
Sangjin) That's right. Our team is divided into frontend and backend developers, and without dedicated planners or designers, we handle planning and design ourselves . As a result, our biggest distinguishing feature is how quickly we make decisions. Usually, development starts after a plan and design are completed. But since the person who plans also builds it, when we think, "Should we try changing it this way?" we can put the idea into action and see the results right away. We start discussions already knowing whether something is technically feasible and how long it will take, which also cuts down on unnecessary communication among us. Another thing is that the whole team is moving in the same direction. We share a clear goal of "making things easier and faster," so planning, design, and development don't happen in silos. When the results of thinking things through and improving them together lead to positive user responses, the whole team feels rewarded.
Heeju) I agree with Sangjin. In a nutshell, our way of working is "build it ourselves, try it ourselves, and fix it right away." Since developers handle every step, including planning and design, when we have an idea, rather than documenting it, it's faster to build it first and show it(laughs). In particular, because our team is made up entirely of developers, we keep making an extra effort not to lose sight of the user's perspective. After we build something, everyone on the team tries it out, and we naturally share candid feedback with one another about any friction we notice from the user's perspective. Since there are no strict role boundaries, the idea that "that's not my job" doesn't really exist. If an idea is good, we put it into practice right away, whoever it comes from.
Q. What challenges do you face in your work?
Heeju) One challenge comes from the fact that we build a tool for creating content. The biggest challenge is that AI can't produce exactly the same result every time. Even with the same script, the results can vary, so it's not easy to establish standards for what counts as "good enough" and build a system that filters out "strange results." Another challenge is meeting users' expectations. They expect results on par with videos filmed and made by people, so consistently meeting those expectations regardless of who uses the service or when is an ongoing task for us. More important than creating one good video is making it possible for anyone to create videos of consistent quality.
Q. On the other hand, what do you enjoy about your work?
Sangjin) The best thing is being able to see the results. Backend work usually happens behind the scenes, so it's hard to see its impact. But with STUDIO PERSO, the features I build appear directly as videos. It's uniquely rewarding as a backend developer to see my code become an actual STUDIO PERSO feature and its results appear in a video.
Heeju) I like being able to gain experience applying AI technology to a real service. New technologies emerge quickly in this field, so there's always something to learn. What's interesting is not just researching them, but incorporating them into a product that real users use. Since our development scope is broad, when we take on a feature, we think through everything from why it's needed to how it should look on screen and work in practice. I feel that this process broadens my perspective as a developer.
Q. What is the most rewarding moment in your work?
Sangjin) Our team has been trying out various ways to use AI lately. When AI features are first applied, the results often fall short of expectations. For example, when generating background images or experimenting with avatars, the initial results can look awkward or differ from what we intended. But as we keep refining prompts, changing the order of processing, and trying different approaches, the results gradually improve. At some point, they reach a level where we think, "Now this is usable." I find that moment really rewarding.
Especially with things that don't have a single right answer, you have to keep experimenting and identify the reasons one by one when you don't get the result you want. So the sense of achievement when something that wasn't working starts to come together feels greater than it does with ordinary feature development. After all the trial and error, it's very rewarding to see the improved results with my own eyes in a video made using the new feature.
Q. Since we're talking about AI, I heard that STUDIO PERSO uses AI to create an LLM Wiki that the whole team uses in its work.
Sangjin) That's right. Our team created an in-house LLM Wiki for the Studio service and uses it in our work. It was frustrating to have to repeat the same information every time we gave Claude a development task.
Our team members know the context of our service well, but every session is a fresh start for Claude. Team members may also have trouble knowing the details of areas they don't handle themselves. Having to explain the same things every time felt inefficient.
So we gathered the knowledge needed for the Studio service in this LLM Wiki. The Wiki's primary readers are LLMs, not people. People manage and verify the knowledge it contains. We organized code logic, context, and feature information for the frontend and server in the Wiki, which helped us identify and address potential issues from the design stage. As a result, we reduced unnecessary communication between backend and frontend teams and naturally shortened development time. Overall, it has greatly improved the team's work efficiency.
Heeju) We work on screen designs in Figma, but since we don't have a designer on the team, we use Figma Make to give instructions through prompts. We also use AI for simple tasks such as checking basic functionality and repetitive QA work. We instruct Claude to operate a web browser directly and test Studio features.
Q. What would you like to work on next?
Heeju) Right now, our service centers on professional avatar models. In the future, we want to open up the opportunity for more people to create their own avatars from just one photo and a short voice recording, then use them to make videos such as video profiles . That way, people who feel uncomfortable speaking naturally in front of a camera or don't have time to film can more easily create videos of themselves speaking in their own face and voice. That's also why we're focused on lowering generation costs and increasing speed.
Sangjin) I want users to have fewer steps to go through when making a video. We want to make the video production automation process faster and more reliable. This aligns with STUDIO PERSO's direction: a service that's easier, faster, and accessible to more people.
![[Itself] The people finding answers in AI by building, trying, and refining: The Perso Studio Team ② Attached image 1](https://framerusercontent.com/images/7uS6asKaDAvX8RwIN58CpXqvU.png)
Q. If STUDIO PERSO were remembered in this world in a sentence or a word, how would you want it to be remembered?
An AI video creation tool that makes it easy to create and deliver a consistent message That's how I'd like it to be remembered. A service that lets anyone easily make videos and deliver their company's story with the same face and voice, no matter how many times they create one.
Q. We're curious about this team's goals! What will STUDIO PERSO look like around the end of 2026?
Heeju) To sum up our team's goal in one sentence: "Let AI handle most of it, while keeping the details editable". The idea is that when you add a document or link, AI drafts everything from the script and screen layout to the avatar's movements, and users only need to edit that draft. That's why we put as much effort into perfecting the editor as we do into automation. When automatically generated content has an error, people need to be able to fix it in three seconds. That's what will make them trust the automation and keep using it.
Sangjin) We want to implement this before the end of 2026. As Heeju said, you provide a document and get a draft that someone can review right away, then simply adjust anything they don't like on the timeline. Ultimately, we want to become more than a simple editor: a system that makes it easy to turn a company's message into a video. A system where you set up your company's unique avatar, voice, and terminology once, and whatever document you enter, the result comes out as if your brand were speaking.
The STUDIO PERSO team is thinking constantly about how to go beyond making individual videos and create an environment where anyone can easily make content featuring an avatar with their own face and voice, while companies can deliver consistent messages through their own distinctive avatars and voices.
The future the team envisions for STUDIO PERSO is a video production automation service where AI handles more of the process and users can easily control only the parts they need to. They're also working on an editing environment where users can quickly review and revise automatically generated results. Even when AI handles much of the process, users need to be able to trust and use the results.
The STUDIO PERSO team is searching for answers that aren't predetermined by building, using, and refining AI themselves . Their ongoing experiments will lead to a day when anyone can easily create all kinds of content with an AI avatar featuring their own likeness. Look forward to what's next from STUDIO PERSO!
![[Itself] The people finding answers in AI by building, trying, and refining: The Perso Studio Team ② Main image](https://framerusercontent.com/images/n2FeGgT1NxinKOob4fFWE1iY.jpg?width=1920&height=1080)
![[ESTsoft] ESTsoft provides Seven Eleven with its intelligent AI avatar “AI Mate” to test its usefulness in convenience store shopping!](https://framerusercontent.com/images/N9N0W8oRGGl0nZWld0DSqKJsdqs.webp?width=1440&height=900)
![[ESTsoft] Real-Time AI Avatar Service Perso Interactive Expands Across Public Services, Building Customized AI Services for Local Governments and Public Institutions](https://framerusercontent.com/images/Iie6Ol3RQYNZcXd8NsouLLX7TT8.webp?width=900&height=502)
