rinna co., ltd. announced on Friday, September 9, the Japanese-specialized “Japanese Stable Diffusion” image generation AI model.
[Image]Image generated by “Japanese Stable Diffusion”
This “Japanese stable broadcast” was developed by additionally learning about 100 million images with Japanese captions to the “stable broadcast” image generation AI model. It is published on the Hugging Face open source community and GitHub.
Stable Diffusion AI open source image generation
“Stable Diffusion” is an AI model that generates images from English character strings developed by Stability AI. It was announced on August 10 and became a hot topic in Japan, sparked by an article published by Takayuki Fukatsu, CXO of note company. After that, it was released as open source, and many services using “Stable Diffusion” appeared in Japan and overseas, such as the LINE app “Oekakibari Good-kun” and the search and browse service “Lexica “.
Approximately 100 million images with Japanese added for additional learning
Regarding the development background of “Japanese Stable Diffusion”, rinna Co., Ltd. stated that “Stable Diffusion”, which uses English, cannot reflect Japanese-specific expressions that are difficult to translate, such as proper nouns, Japanese English, and onomatopoeia. . is an image of the English-speaking world, and an image that strongly reflects the culture of the English-speaking world is generated. Therefore, in order to support Japanese, they developed “Japanese Stable Diffusion” through additional learning using around 100 million images with Japanese captions. In addition, rinna Co., Ltd. “released the model early to keep pace with advancements in the English-speaking world in the field of AI which is advancing significantly.” “Japanese Stable Diffusion” is also implemented on the AI character SNS “Chararu” operated by rinna Inc. and the official AI Rinna Twitter. See the press release for more details.