The creative community responds to Lensa AI, the “magic avatar” generator
It’s soared to the top of the Apple download charts, with portraits being shared by everyone from your best friend to Chance the Rapper. But growing grievances from artists are placing a question mark over the app.
- Liz Gorny
- 7 December 2022
Lensa is not a new app. It was launched in 2018 by Prisma Labs; its meteoric rise in popularity has come since a new update to the app: Magic Avatars, which is what enables the creation of the “alter ego” selfies you’ve seen across social media. Lensa works by using a copy of the Stable Diffusion model – a deep learning, text-to-image model released this year – and requires 10-20 images of the user to generate avatars.
“As soon as the avatars are generated, the user’s photos and the associated model are erased permanently from our servers. And the process would start over again for the next request.”
Then there’s the conversation around how Lensa exactly uses work from other artists, with many creatives across Twitter asking users to look into how Lensa works. Artist and illustrator Meg Rae, for example, stated recently that Stable Diffusion images are created using “stolen art”. The conversations around artist ownership and AI models is ongoing – there is now even an online tool to let creatives find out if their images have been used to train AI models like Stable Diffusion. On its website Spawning explains how it created the tool to enable artists to opt into or opt out of the training of large AI models. In August, concept artist RJ Palmer shared a selection of artists “advertised as styles” by Stable Diffusion.
In response to Lensa Magic Avatars specifically, many are also pointing to visual examples of artist work within the portraits. Graphic designer Lauryn Ipsum has shared a selection of Lensa portraits on Twitter, in which “the mangled remains of an artist’s signature is still visible”.
In Prisma Labs’ recent statement explaining how AI generates images, the company states: “AI is capable of rapidly analysing and learning from large sets of data, but it does not have the same level of attention and appreciation for art as a human being. The AI learns to recognise the connections between the images and their descriptions, not the artworks. This way the model develops operational principles that can be applied to content generation. Hence the outputs can’t be described as exact replicas of any particular artwork.”
In conversation with The Guardian, Lensa disputed the claim that portraits included traces of signatures on 11 December. A spokesperson from the app stated: “The blurred silhouettes of visual elements [that one may perceive as signatures] you see in the outputs do not utilise any existing language; often, they don’t feature any letters.” The spokesperson continues: “They do not represent the remains of existing artists’ signatures either.”
Smalllike: Selfie, from The Noun Project
About the Author
Liz (she/they) joined It’s Nice That as news writer in December 2021. After graduating in Film from The University of Bristol, she worked freelance, writing for independent publications such as Little White Lies, INDIE magazine and design studio Evermade.