Apps

It’s way too easy to trick Lensa AI into making NSFW images

Comment

Lensa AI can be tricked into leaving very little to the imagination. Illustration of a woman's bare shoulder...
Image Credits: Lensa AI

Lensa has been climbing the app store hit lists with its avatar-generating AI that is making artists wave the red flag. Now there’s another reason to fly the flag: As it turns out, it’s possible — and way too easy — to use the platform to generate non-consensual soft porn.

TechCrunch has seen photo sets generated with the Lensa app that include images with breasts and nipples clearly visible in the images with faces of recognizable people. It seemed like the kind of thing that shouldn’t have been possible, so we decided to try it ourselves. To verify that Lensa will create the images it perhaps shouldn’t, we created two sets of Lensa avatars:

  • One set, based on 15 photos of a well-known actor.
  • Another set, based on the same 15 photos, but with an additional set of five photos added of the same actor’s face, Photoshopped onto topless models.

The first set of images was in line with the AI avatars we’ve seen Lensa generate in the past. The second set, however, was a lot spicier than we were expecting. It turns out the AI takes those Photoshopped images as permission to go wild, and it appears it disables an NSFW filter. Out of the 100-image set, 11 were topless photos of higher quality (or, at least with higher stylistic consistency) than the poorly done edited topless photos the AI was given as input.

AI is getting better at generating porn. We might not be prepared for the consequences.

Generating saucy images of celebrities is one thing, and as illustrated by the source images we were able to find, there has long been people on the internet who are willing to collage some images together in Photoshop. Just because it’s common doesn’t make it right — in point of fact, celebrities absolutely deserve their privacy and should definitely not be made victims of non-consensual sexualized depictions. But so far, getting those to look realistic takes a lot of skill with photo editing tools along with hours, if not days, of work.

The big turning point, and the ethical nightmare, is the ease with which you can create near-photorealistic AI-generated art images by the hundreds without any tools other than a smartphone, an app and a few dollars.

The ease with which you can create images of anyone you can imagine (or, at least, anyone you have a handful of photos of), is terrifying. Adding NSFW content into the mix, and we are careening into some pretty murky territory very quickly: your friends or some random person you met in a bar and exchanged Facebook friend status with may not have given consent to someone generating soft-core porn of them.

It appears that if you have 10-15 “real” photos of a person and are willing to take the time to photoshop a handful of fakes, Lensa will gladly churn out a number of problematic images.

AI art generators are already churning out pornography by the thousands of images, exemplified by the likes of Unstable Diffusion and others. These platforms, and the unfettered proliferation of other so-called “deepfake” platforms, are turning into an ethical nightmare, are prompting the U.K. government to push for laws criminalizing the dissemination of non-consensual nude photos. This seems like a very good idea, but the internet is a hard-to-govern place at the best of times, and we’re collectively facing a wall of legal, moral and ethical quandaries.


UPDATE: The Prisma Labs team replied to our concerns. The company highlights that if you specifically provoke the AI into generating NSFW images, it might, but that it is implementing filters to prevent this from happening accidentally. The jury is still out as to whether this will actually help people who are the victim of this sort of thing without their consent. 

Prisma Labs, maker of Lensa AI, says it is working to prevent accidental generation of nudes

More TechCrunch

Public market investors have a large variety of infrastructure and software that helps them keep track of, analyze and manage their investments, but that’s not the case for investors in…

bunch raises $15.5M for its platform that simplifies investment management for VCs

India’s Jio has partnered with Taiwanese semiconductor giant MediaTek to launch its 4G smart dashboards for electric two-wheelers.

Jio partners with Taiwan’s MediaTek to tap into two-wheeler EV market

A hacker claims to be selling data relating to thousands of current and former employees of India’s Piramal Group.

Hacker claims theft of Piramal Group’s employee data

CRED, an Indian fintech startup, has rolled out a new feature that will help its customers manage and gain deeper insights into their cash flow, as the startup seeks to…

CRED launches personal finance manager for India’s affluent

A powerful new video-generating AI model became widely available today — but there’s a catch: The model appears to be censoring topics deemed too politically sensitive by the government in…

A new Chinese video-generating model appears to be censoring politically sensitive topics

Our growth as a civilization is tightly coupled to our ability to sufficiently generate ever-increasing amounts of electricity. Could the same be true in space?  Star Catcher Industries, a startup…

Star Catcher wants to build a space power grid to supercharge orbital industry

For frontier AI models, when it rains, it pours. Mistral released a fresh new flagship model on Wednesday, Large 2, which it claims to be on par with the latest…

Mistral’s Large 2 is its answer to Meta and OpenAI’s latest models

Researchers at MIT CSAIL this week are showcasing a new method for training home robots in simulation.

Researchers are training home robots in simulations based on iPhone scans

Apple announced on Wednesday that Apple Maps is now available on the web via a public beta, which means you can now access the service directly from your browser. The…

Apple Maps launches on the web to challenge Google Maps

AltStore, an alternative app store, has launched its first batch of third-party iOS apps in the European Union. The rollout comes a few months after the company launched an updated…

Alternative app store AltStore PAL adds third-party iOS apps in wake of EU Apple ruling

Microsoft this afternoon previewed its answer to Google’s AI-powered search experiences: Bing generative search. Available for only a “small percentage” of users at the moment, Bing generative search, underpinned by…

Bing previews its answer to Google’s AI Overviews

Hiya, folks, welcome to TechCrunch’s regular AI newsletter. Last Sunday, President Joe Biden announced that he no longer plans to seek reelection, instead offering his “full endorsement” of VP Kamala…

This Week in AI: How Kamala Harris might regulate AI

But the fate of many generative AI businesses — even the best-funded ones — looks murky.

VCs are still pouring billions into generative AI startups

Thousands of stories have been written about former NFL quarterback and civil rights activist Colin Kaepernick. If anyone knows a thing or two about losing control of your own narrative,…

Colin Kaepernick lost control of his story. Now he wants to help creators own theirs

Several people who received the CrowdStrike offer found that the gift card didn’t work, while others got an error saying the voucher had been canceled.

CrowdStrike offers a $10 apology gift card to say sorry for outage

TikTok Lite, a low-bandwidth version of the video platform popular across Africa, Asia and Latin America, is exposing users to harmful content because of its lack of safety features compared…

TikTok Lite exposes users to harmful content, say Mozilla researchers

If the models continue eating each other’s data, perhaps without even knowing it, they’ll progressively get weirder and dumber until they collapse.

‘Model collapse’: Scientists warn against letting AI eat its own tail

Astranis has fully funded its next-generation satellite program, called Omega, after closing its $200 million Series D round, the company said Wednesday.  “This next satellite is really the milestone into…

Astranis is set to build Omega constellation after $200M Series D

Reworkd’s founders went viral on GitHub last year with AgentGPT, a free tool to build AI agents that acquired more than 100,000 daily users in a week. This earned them…

After AgentGPT’s success, Reworkd pivots to web-scraping AI agents

We’re so excited to announce that we’ve added a dedicated AI Stage presented by Google Cloud to TechCrunch Disrupt 2024. It joins Fintech, SaaS and Space as the other industry-focused…

Announcing the agenda for the AI Stage at TechCrunch Disrupt 2024

The firm has numerous legs to it, ranging from a venture studio to standard funds, where it does everything from co-founding companies to deploying capital.

CityRock launches second fund to back founders from diverse backgrounds

Since launching xAI last year, Elon Musk has been using X as a sandbox to test some of the Grok model’s AI capabilities. Beyond the basic chatbot, X uses the…

X launches underwhelming Grok-powered ‘More About This Account’ feature

Lakera, a Swiss startup that’s building technology to protect generative AI applications from malicious prompts and other threats, has raised $20 million in a Series A round led by European…

Lakera, which protects enterprises from LLM vulnerabilities, raises $20M

Alongside a slew of announcements for Play — such as AI-powered app comparisons and a feature that bundles similar apps — Google has introduced new “Curated Spaces,” hubs dedicated to…

Google Play gets ‘Comics’ feature for manga readers in Japan

Farmers have got to do something about pests. But nobody really likes the idea of using more chemical pesticides. Thomas Laurent’s company, Micropep, thinks the answer might already be in…

Micropep taps tiny proteins to make pesticides safer

Play Store is getting AI-powered app comparisons, automatically organized categories for similar apps, dedicated hubs for content, data personalization controls, support for playing multiple mobile games on PCs, and more…

Google adds AI-powered comparisons, collections and more data controls to Play Store

Vanta, a trust management platform that helps businesses automate much of their security and compliance processes, today announced that it has raised a $150 million Series C funding round led…

Vanta raises $150M Series C, now valued at $2.45B

The Overture Maps Foundation is today releasing data sets for 2.3B building “footprints” globally, 54M notable places of interest, a visual overlay of “boundaries,” and land and water features such…

Backed by Microsoft, AWS and Meta, the Overture Maps Foundation launches its first open map datasets

The startup is not disclosing its valuation, but sources close to the company say the figure is just under $400 million post-money.

Dazz snaps up $50M for AI-based, automated cloud security remediation

The outcome of the Spanish authority’s probe could take up to two years to complete, and leave Apple on the hook for fines in the billions.

Apple’s App Store hit with antitrust probe in Spain