OpenLLM-Vtuber
I’d like to create an avatar of Shorekeeper using the github Item “OpenLLM-Vtuber”. This blog is to trace my progress.
2025/3/11
I cloned the repository and ran the code. Using my deepseek API from “火山方舟”, I successed in starting basic communication on local server with suiku, the default avatar in the Item. I gave her the prompt of shorekeeper.
Future plans:
- Create a new avatar of shorekeeper using Live2D. But it’s quite “Art”, maybe hard for a programmer like me to do it.
- Replace the default AudioEngine with my own. I think it’s possible, but I need to learn more about the AudioEngine.
- Use my local deepseek-r1-distill-Qwen-14B to run the avatar. Whether she will be smarter or not remains to be seen.
2025/3/14
I found a great AI audio creater GPT-Sovits. I generated an example:
2025/4/21
I returned to the Item after a long time. This time I realized using the live2d model of 秧秧 instead of the default one. I also used a prompt of “Queen”. See here:
next step:
- Use my local deepseek-r1-distill-Qwen-14B to run the avatar.
- Find a well-drowned avatar.
- Replace the audio engin.
