【Stable Diffusion Online】商用利用ＯＫの画像生成AI『Stable Diffusion』

The speaker introduces Stable Diffusion, an open-source image generation AI that can be run on moderately high-performance computers. Unlike cloud-based AIs like DALL-E and Midjourney, Stable Diffusion allows users to modify the AI's program on their own computers. The presenter mentions that using a web-based interface is easier than running it directly. The script explains that Stable Diffusion can generate realistic images from any text input, offering a creative freedom that could empower billions to produce art within seconds. The online platform allows users to create beautiful art for free, with sections like Playground, Features, and FAQ. The platform is easy to use and can be utilized to create realistic images, with an open license that permits both personal and commercial use.

The speaker shares their practical experience with Stable Diffusion, noting that it can be time-consuming. They have pre-experimented with various features such as 'Candies' and 'Head' transformations, and interactive fake scribbles. Despite the complex terminology like 'Segmentation Depth Normal Map', the speaker experimented with all the features. They demonstrate how uploading an original image and applying different transformations can result in varied outputs, such as changing the pose of a person or altering the clothing. The speaker also mentions the 'ControlNet' feature, which is described as a game-changer for AI image generation, allowing additional controls over the diffusion model. However, they caution that due to high user numbers, server issues and errors might occur, advising users to try again if they encounter errors. The speaker concludes that while the quality might not be the best compared to other AI image generation tools, Stable Diffusion is enjoyable for casual use and allows for both personal and commercial utilization of created images.