NovelAIDiffusion storytelling, image generation

NovelAIDiffusion is an AI model that generates images from text prompts, and prose from text prompts, to generate image and text-based stories. It can be used to generate characters and plotlines in a wide range of genres, including in the style of Arthur Conan Doyle and Edgar Allan Poe. 

Developed and operated by US-based company Anlatan based on data from Danbooru, a two-dimensional illustration sharing website, and GPT large language models, NovelAI was first launched in June 2021 as an AI storytelling service. Its Stable Diffusion-powered 'NovelAIDiffusion' image generation capability was added in October 2021.

NovelAI has been enthusiastically adopted in Japan, where it is known for producing high quality anime, manga, and hentai imagery. It is unclear whether, to what extent, and how it is being used in other countries and regions. 

Risks and harms 🛑

NovelAIDiffusion has been criticised for copyright abuse, potentially damaging human creativity, and resulting in the loss of jobs.

Copyright ethics

• The launch of NovelAIDiffusion resulted in artists, illustrators and others complaining that data used to build the model had been scraped from the web without the knowledge of permission of the people who created the images in the first place. 

• This has been seen as particularly objectionable as NovelAI is subscription-based, and its system has not been made available to others - unlike Stable Diffusion - even if it is not technically illegal to scrape others' content and artwork in Japan or the USA.

• Further complicating matters is the fact that NovelAI's terms of service says content generated by the system belongs to the user. Unsurprisingly, artists are worried that they will not be compensated for their work, however close it is to art generated by Novel AI's system.

• Lawyer Kazuyasu Shiraishi noted to Rest of the World that copyright issues only arise in Japan if the output is exactly the same, or very close to, the images on which the model is trained.

Genshin Impact art theft

• A few weeks after NovelAIDiffusion had launched, Kotaku reported that popular Korean-language artist and Twitch streamer AT livestreamed an 11-hour sketch session in which he constructed an image of Raiden Shogun, a popular character from the video game Genshin Impact. 

• The image was stolen, 'completed' using NovelAI, and shared on Twitter by another user under the handle @Musaishh before AT could finish it. The thief then demanded credit from AT. The ensuing outcry persuaded the thief to delete their Twitter account and disappear.

• The increasing popularity of NovelAI, and incidents such as Genshin Impact artwork theft, led to a number of high-profile Japanese websites and communities insisting on AI-generated art being labelled as such.

News, commentary, analysis 🗞️

• https://dataconomy.com/2022/10/novelai-novelaidiffusion/

• https://news.artnet.com/art-world/artificial-intelligence-illustration-spawning-2195919

• https://screenshot-media.com/technology/ai/ai-art-thief/

• https://kotaku.com/genshin-impact-fanart-ai-generated-stolen-twitch-1849655704

• https://www.sohu.com/a/591569563_104421

• https://ricedigital.co.uk/dont-use-ai-to-disrespect-artists/

• https://restofworld.org/2022/ai-backlash-anime-artists/