Rampant Scraper Midjourney Bans Rival AI Firm For Scraping MJ images

March 13, 2024

[adinserter block=”1″]

Midjourney blamed a 24-hour outage on what it describes as “bot-net-like” activity by an image-scraping employee supposedly working for rival Stability AI.

These suspicions then prompted the AI image-rendering company to indefinitely ban all employees of Stability AI from using the Midjourney service.

According to Midjourney, the suspected Stability employee was trying to scrape prompt image pairs in bulk to such a degree that it caused a Midjourney service outage on March 2.

The announcement was tweeted by Midjourney advocate Nick St. Pierre after being posted by Midjourney’s official Discord channel.

For those of you who still aren’t aware of how AI image rendering works, prompts are the pieces of descriptive/instructive text given to generative AI models by human users for the sake of having the AIs generate images from them.

If someone were to obtain connected pairs of prompts and the images created from them by an AI, they could use these to fine-tune the training of their own AI systems.

Stability AI’s image-generating AI is called Stable Diffusion and up until recently many considered it to be the best AI rendering program for creating high quality images easily.

However, more recently, Midjourney released version 6 (Midjourney V6) of its AI and it seems to create notably better results than any other major consumer system of this kind.

Thus, if Midjourney is widely regarded by many users as being the most capable of the image-rendering AI’s out there today, scraping its rendering/prompt pairs for training could make sense for a rival AI firm.

The bot activity described by MJ happened on March 2 around midnight and caused the image-generating services to go down for users during roughly 24 hours.

Midjourney staff then apparently managed to link several paid accounts to their service with a data team employee of Stability AI.

Midjourney claimed that these accounts were trying to snatch up prompt and image pairs on a massive (scraping level) scale.

Shortly after, all Stability AI employees were banned from the service indefinitely.

In MJ office hours they just said someone at Stability AI was trying to grab all the prompt and image pairs in the middle of a night on Saturday and brought down their service.

MJ is banning all of the stabilityAI employees from Midjourney immediately

This is breaking now

— Nick St. Pierre (@nickfloats) March 6, 2024

MJ also declared a new policy: any aggressive automation from an account that results in the MJ AI service failing will lead to bans for all employees of the company holding that account.

Based on the evidence so far, it seems reasonable to say that the Stability AI employee was practicing some pretty terrible operational secrecy if they were using such easily traced accounts to scrape image/prompt pairs at scale.

Following the ban and policy changes, Stability responded to Midjoureny’s claims through a tweet in which its CEO Emad Mostaque claimed that he was looking into whatever happened and promised that it wasn’t intentional.

If anyone did do this on team (have asked, will dig, also happy if MJ reaches out direct) its not great but obviously not a DDoS attack but unintentional.

Certainly not instructed by us/Stability AI tho, really happy with our dataset & the augmentations we have on that.

— Emad acc/acc (@EMostaque) March 6, 2024

The rival firm’s CEO also added that Midjourney should reach out to him directly.

Responding, Midjourney’s CEO David Holz tweeted, “sent you some information to help with your internal investigation”.

In text messages to staff at tech site Ars Technica, which covers this story too, Mostaque claimed,

“We checked and there were no images scraped there, there was a bot run by a team member that was collecting prompts for a personal project though. We aren’t sure how that would cause a gallery site outage but are sorry if it did, Midjourney is great.”

The stability CEO also threw in that the accusations don’t make sense since his company doesn’t need anything from Midjourney. According to him, Stable Diffusion’s latest version, SD3, outperforms Midjourney in any case.

There are generative AI users of all kinds who might beg to differ on that claim, but that’s another debate.

Mostaque added in messages to Ars, “We only scrape stuff that has proper robots.txt and is permissive,” contrasting his company’s data collection techniques with those of MJ.

Mostaque further added that his company’s relationship with Midjourney hasn’t been too bad in general. “No real overlap, we get on fine though,”

Interestingly, he also claimed that he had funded Midjoureny when it was getting started by giving them a cash grant to cover the costs of Nvidia GPUs for their AI data-crunching needs.

Midjourney obviously suffered something that upset its operations and its senior staff and seems to genuinely suspect an employee of Stability of being at fault.

However, it’s hard to say if the Stability CEO is being honest that he hadn’t played any knowing part in the intrusive scraping. An employee could also have simply done this on their own wayward initiative, though it seems a bit unlikely.

A further irony in all of this is that Midjourney itself built its entire business on the rampant scraping of images owned and created by third parties, without their permission or compensation.

Now it turns out that it doesn’t like getting a dose of its own bitter medicine (though admittedly to our knowledge MJ’s own scraping never caused anyone to suffer a full blown systems outage.)

Midjourney and other AI rendering companies (Stability AI included) are in any case all getting heat for some of their bot practices.

Millions of artistic works by thousands of artists and photographers have been used without their pre-approval to train the AIs of these companies and many creators have become furious with this.

As one artist tweeted, “Words can’t describe how dehumanizing it is to see my name used 20,000+ times in MidJourney,”

She added, “My life’s work and who I am—reduced to meaningless fodder for a commercial image slot machine.”

Seeing just how quickly MJ and the other AIs can render artistic imagery in seconds due to vast training sets of art that human hands took hours or days to produce, one piece at a time can easily be a painful experience for any artist.