This article was written by a human and proofread/translated by Gemini 3.1 Pro.
Released VTuberKitForYMM4
I recently released VTuberKitForYMM4. I tried to see if I could raise the quality of a plugin I made over a year ago to a product level using the latest AI, but even with that, it took about a month to finish. Combined with manual work, it might have taken around 4 months in total.
Live2D Licensing Caveats and My Response
After publishing, I encountered an issue. I use the Live2D Cubism SDK Release License (Publishing License Agreement), but as it turned out, in the case of an “Extensible Application” that can load custom Live2D models, even individuals or small businesses (annual sales under 10 million JPY) need to sign a contract, and distributing exe or dll files requires publishing permission. This was pointed out to me in an issue, so I hurriedly deleted the dll files. (When I temporarily changed the repository to private to fix this, all my GitHub stars blew away and went back to 0…)
I remembered checking this more than a year ago, but while being careful not to upload the SDK itself, I made the mistake of uploading it to the release assets. I sincerely apologize to anyone who downloaded it.
After directly confirming with Live2D Inc., I received the following answer:
- Distributing the repository on GitHub “excluding the SDK itself and the final release output” is not a problem.
- Content such as videos created using the Live2D Cubism SDK may require users to sign a separate contract depending on their scale (not required for individuals or small businesses; required for medium businesses or larger, meaning annual sales of 10 million JPY or more).
- In the case of an “extensible application that can load Live2D models” using the Live2D Cubism SDK, even individuals or small businesses may need a license, so consultation is required.
So, there is almost no problem for individual users to output videos, so please use it with peace of mind.
By the way, I remembered there was a free version of VTubeStudio and looked it up, but as mentioned in this tweet, the free use ended in June 2025. When I looked further into VTubeStudio, it said “A contract with Live2D Inc. is required if sales exceed 20 million JPY,” so it seems like a different contract structure.
Development Effort and Struggles with WPF
Here’s some talk regarding the development of VTuberKitForYMM4. This project can be roughly divided into three parts: “Live2D,” “YMM4,” and the “WPF UI.” The ratio of effort felt during development using current AI is roughly: Live2D: 0.5, YMM4: 0.5, WPF: 9 In other words, I struggled the most with implementing the UI (WPF).
I know it is Microsoft’s great library, but it seems AI still only “understands it as strings,” and I kept having bugs with WPF event notifications and similar things (there was the frustration of the AI not understanding even when I showed it screenshots). However, since AI is evolving quickly this year as well, I expect that by the time I build the next UI, it should be possible to implement more smoothly.
From here on, it’s just a diary
AviUtl2 is still surprisingly popular, and I feel YMM4’s name recognition is still somewhat low. (Even now in 2026, there are people who believe that “AviUtl is running behind the scenes.”) A post flowed through X (formerly Twitter) saying, “The first impression is very important for humans, and it takes 10 interactions to overturn it.”
What I mean to say is that by publishing it for free and unexpectedly violating a license, I am concerned that a “dangerous image” might have been attached to me. Winning back trust is hard, and doing this is essentially unpaid work. It’s weird to ask Live2D Inc., who does this for business, “I want to publish this for free, so please sign a contract with me,” and fundamentally, a salaried worker working for free is not a good idea. So honestly, “signing contracts and taking on work for free” is bizarre; OSS activity is ultimately a passion project for a hobby.
Generative AI Environment and the Future
Currently, I’m using AI like Copilot and the Gemini family through the Open University of Japan, but they have suddenly been heavily restricted (nerfed), and the situation where I can’t freely use generative AI is quickly approaching. I feel there isn’t much more I can do. (Also, I’ve been allowed to use Codex for a month via ChatGPT’s free period, but that will also end in March.)
If I find myself in a situation where I can only utilize Gemini, I thought game development (Roblox) might be suitable, since “high code quality is not required, but it shines in the multimodal and 3D domain.” However, Roblox has also seen a 1/10th drop in active users due to impacts like the chat feature being prohibited. (Even though I haven’t even started anything yet…)
The Common Charm of YouTube and Roblox
That said, I think YouTube and Roblox have similar aspects. What’s especially great is that you don’t need exhausting tasks like “advertising every day” on X. (Although you might need ads within Roblox itself).
Therefore, I feel focusing on Roblox development from now on is the best idea. There are other reasons, the biggest being “adults hate it.” I still like playing games myself (like Slay the Spire), but I definitely played much more when I was a kid. That’s exactly why working on Roblox, which appears attractive to kids, seems like a good choice.
For now, I’ll keep clicking “Like” on YMM4 plugin creators’ posts, use AI conversations as an echo chamber, and move forward with Roblox development.