• Fri. Sep 29th, 2023



The enthusiast involved Stable Diffusion neural network for selfie processing


Jan 31, 2023

The enthusiast involved Stable Diffusion neural network for selfie processing

German enthusiast Fabian Shtelzer used online tools based on the popular Stable Diffusion neural network to process his selfie in different styles. He popularized such algorithmic systems and made them viral, writes Motherboard.

The Stertzer involved programs that several times make user settings to the image generator in the text.

He created a selfie in various styles of painting, films genres and in his idea like a photo from Woodstock or in the image of a woman.

Stable Diffusion from Stable Ai startup is one of many JI systems that generates images on request. The developers have trained an algorithm for billions of annotated and marked pictures.

The neural network can be trained to associate specific words and phrases with certain types of images, aesthetics, places or objects. Thanks to this, websites appeared that allow users to “regulate” the model and add their own embedding.


The creators of the DRAWANYONE service claim that their tool is able to process photos on five source pictures in just an hour. Several people demonstrated their AI-Selphi in the thread under the post-by.

According to the enthusiast, he was struck by the quality of the system, as it required only a few reference images to “introduce oneself” into the model.

“The selfies loaded me are a little the same, so I will re -train the tool on a more diverse set of poses and expressions,” he said.

The Shtelzer also engages in other AI projects, including the development of a full-length multi-tower film “Salt”. It involves algorithms like Stable Diffusion, Midjourney and Dall-E 2 to create frames and sound generation model, including Synthesia and Murf. The neural network GPT-3 is responsible for writing the script.

As each fragment of the film appears, viewers can vote for one of the continuation options.

“[AI systems] are at least as great as the opening of a photo or movie, and if we turn on language models like GPT-3, then, probably, this can be compared with the invention of the printing press. The influence on the media, the culture and fabric of reality will become quite deep, ”the Shtelzer said.

The enthusiast noted that the decision on the publication of the source code adopted by Stability AI will make the algorithm safer than closed corporate projects.

“Attackers also use such systems. However, fewer users who can correctly use the tool work with similar projects, ”he added.

Recall that in October the enthusiast involved Stable Diffusion to create constantly changing virtual worlds.

In the same month, the AI-Researcher taught the neural network to turn “anyone” into Pokemon.

In August, Stability AI provided more than 1000 developers with access to the image generator in the text of Stable Diffusion, and later opened it for a wide audience.

Subscribe to FORKLOG news at Telegram: Forklog AI – all news from the world of AI!

Leave a Reply

Your email address will not be published. Required fields are marked *