Wix-owned online art community DeviantArt announced DreamUp, an AI-based text-to-image generation service, on Friday. Similar to other projects that make use of existing machine learning models to power their services, DreamUp will use Stable Diffusion to power this new facility.
It does come at a cost, however. While new users will get five free prompts on the DreamUp page, generating further art will require upgrading to a Core plan on the website. There are currently four subscription tiers offering up to 300 free monthly prompts starting at $3.95 a month.
|DreamUp Features||Core||Core+||Pro||Pro +|
|Free Monthly Prompts||50||100||200||300|
|Images Per Prompt||3||3||4||4|
|Priority in Queue||High||High||High||High|
|Cost Per Additional Prompt||20 Points||15 points||10 Points||5 Points|
These AI models are trained on images scraped from the internet and have been a topic of controversy in the past. Stable Diffusion is among the most controversial of these, with its image-scraping habits causing problems with artists challenging the ethics of AI image generation. While DeviantArt has allowed artists to opt out of the training process, it doesn’t completely keep their images from eventually being used.
Adding fuel to fire, artists need to manually opt-out of their existing deviations (artworks on DeviantArt) from the AI data set. The site has introduced a ‘noai’ flag that keeps creations from being included in third-party image datasets, but all images, including new uploads, are opted in by default. Whether or not third-party image scrapers will adhere to this flag remains to be seen.
Artists that don’t want their images being used in the dataset for DreamUp will also fill out a form which requires a human review. All these factors combined have led to significant pushback from the artists on the site, with some threatening to delete their work and leave the platform for good.
All this backlash, however, seems to have had some effect on the company’s announcement. In an update to their original announcement, DeviantArt added that all deviations are automatically labelled as not authorised for use in AI datasets and has further clarified its stance on AI-generated imagery.
With DALL-E’s API being launched for public beta and stock photo sites Shutterstock and Getty Images announcing similar features — partnering with OpenAI and BRIA, respectively, artists and photographers alike are lobbying to prevent their images from being used to train AI models which from the looks of it might soon endanger their jobs. Shutterstock and Getty Images have announced artist royalties, but DeviantArt is yet to do so.
In the News: Binance announces recovery fund for “otherwise strong” projects
This made me thought that, the songs on streaming services might be used as samples to ‘educate’ AI. It’s likely.
This would largely depend on what exactly the ‘education’ is about. One possible use case could be training an AI model to quickly transcribe speech into text or how to listen to otherwise regular-sounding voices and speech patterns, sure you could use songs.