https://arstechnica.com/ai/2023/08/google-record-labels-working-on-deal-covering-musical-deepfakes/

"Google and Universal Music are in talks to license artists� melodies and voices for songs generated by artificial intelligence as the music business tries to monetize one of its biggest threats....Warner Music, the third-largest music label, has also been talking to Google about a product, said a person familiar with the matter."

I hope that in time we lose the cavalier attitude about it being OK to steal copyrighted, human-created content because you can't do fun AI things without it..

Edited Oct 20 to add a story about Universal Music suing Anthropic for using copyrighted lyrics in its clone of ChatGPT (Claude):

https://arstechnica.com/tech-policy/2023/10/universal-music-sues-ai-start-up-anthropic-for-scraping-song-lyrics/

"In addition to regurgitating lyrics, Claude responded to prompts asking for writing in the style of popular musicians with unlicensed lyrics, the music companies alleged. �When we asked the AI model to write a piece of short fiction in the style of Louis Armstrong, it uses the lyrics for �What a Wonderful World,�� the companies said in the filing."

"�Publishers embrace innovation and recognize the great promise of AI when used ethically and responsibly. But Anthropic violates these principles on a systematic and widespread basis,� the music groups said in Wednesday�s lawsuit."

I've spent way too much time looking at AdobeStock's fast-growing genAI collection and regularly have that flash of recognition for a face - a sense that I'm sure I've seen that person before.

I've tracked a couple of these down using AdobeStock's excellent "Find Similar feature and it's as if there are a handful of models for each age/ethnicity group and everyone's using this tiny set of people. Their pictures are everywhere!

I don't think this is copying another's images; I think it's the limits of the engines contributors are using to create content. There are now 13,860,209 items in the genAI collection (obviously not all people) and apparently a limited set of pretty faces to draw on. Each one of my examples is from a different contributor.

Here's a selection from 24,355 similar images to the top left item I saw this afternoon (click to see full size)



You can see the entire set of images here.

I'm sure people will point out that photographers can share models in the real world too, but the issue is scale - AI can just pump out near-endless quantities of this stuff. From a buyer's point of view, I don't want every other company's ad to look so similar, and that's going to become extraordinarily hard.

The guy's impossibly fit, tanned and handsome - and multi-talented, well traveled and perennially happy. Ideal stock model if he could ration his appearances just a little

NY Times article (paywall) about fake travel guides for sale on amazon. One book used as an example had an AI generated author photo - they pointed out the tell-tale signs.

https://www.nytimes.com/2023/08/05/travel/amazon-guidebooks-artificial-intelligence.html

amazon should do better in weeding this print-on-demand fakery out, but the scale of the problem when there are bucketloads of these types of fakes being generated makes it hard even if you are being dilligent.

"Mike Steves�s author photo shows anomalies consistent with its having been created by A.I., including unnatural elements on or near the ears (in this case, a partially formed earring) distorted clothing and a blurry and abstract background."

The story is overall about the proliferation of AI-generated largely useless guides with fake "rave" reviews:

"a new form of travel scam: shoddy guidebooks that appear to be compiled with the help of generative artificial intelligence, self-published and bolstered by sham reviews, that have proliferated in recent months on Amazon. The books are the result of a swirling mix of modern tools: A.I. apps that can produce text and fake portraits; websites with a seemingly endless array of stock photos and graphics; self-publishing platforms � like Amazon�s Kindle Direct Publishing � with few guardrails against the use of A.I.; and the ability to solicit, purchase and post phony online reviews, which runs counter to Amazon�s policies and may soon face increased regulation from the Federal Trade Commission."

The reason I would be nervous about that is it potentially signals lower earnings overall.

If you used to earn $250 a month, a $100 payout isn't an issue. When earnings drop to $99 a month, you're unhappy for two reasons. So they drop the payout to $50 so at least you get paid every month. Repeat the cycle as earnings drop to $49 a month etc. etc.

Especially if they see a future with lots of contributors getting data licensing only payouts - ones that come infrequently - having a lower threshold means they don't get people having to wait 6 months (or whatever the data licensing contributor fund interval is) to get paid.

Under the headline AI Lovefest Fuels Massive Bets On Meta, Tesla, Google And Beyond...

"Companies as disparate as Palantir (PLTR), Adobe (ADBE) and Workday (WDAY) all have tapped into the AI zeitgeist while sparking interest among leading money managers....Just last month, Adobe expanded the availability of its generative AI tool to support text prompts in over 100 languages. On July 6, ADBE stock was featured as the IBD Stock Of The Day as the company took aims to monetize generative AI.

https://www.investors.com/etfs-and-funds/mutual-funds/meta-stock-tesla-google-panw-lead-ai-stocks-to-watch-as-best-mutual-funds-fuel-artificial-intelligence-lovefest/?src=A00220

Also this morning:

https://www.fool.com/investing/2023/08/04/3-ai-stocks-that-are-screaming-buys-in-august/

"Over the past decade, Adobe transformed all of its flagship desktop applications -- including Photoshop, Premiere Pro, and Illustrator -- into subscription-based cloud services...

Earlier this year, Adobe expanded its older Sensei AI and machine-learning framework with a new generative AI platform called Firefly. The integration of Firefly into its Creative Cloud will streamline the production of digital media by enabling its users to create images, videos, and digital models with simple text-based prompts.

...These sweeping upgrades could spark another multiyear growth cycle for Adobe as the AI market expands."

On the subject of SS adding 300k contributors in Q2 - what they said was "...the largest uptick we have ever had in contributors in the 20-plus year history of Shutterstock"...

I had a look at new uploads on SS and saw a lot of work I'd bet was genAI created. Here's one example of a new portfolio with a small number of items, almost all labeled "oil painting" (a couple are "acrylic"). They don't appear to be oil paint on canvas to my eye, and hands (Queen Victoria and the dancing girl in orange) look very genAI.

https://www.shutterstock.com/g/Nida+Artist?sort=newest

Perhaps they are really naive enough to believe they aren't getting any genAI content uploaded by contributors. Perhaps they want it, but want plausible deniability in case they're sued.

Look at these fingers - SS has contributor-uploaded genAI content in the library...

Thanks for providing that access.

I thought it was interesting that they will now be providing details on money earned from their (contributors') data:

"In addition, beginning today we will be breaking out the revenue derived from our data engine. As a reminder, our data offering is driven by our ability to monetize our unique repository of metadata which customers license for model training purposes. This includes our multi-year partnerships with large tech companies such as OpenAI, LG, Meta, NVIDIA and now our newly announced transformational partnership with Google. With this new disclosure, we believe that investors will have greater visibility into the scale and trajectory of a major growth driver of our business."

I've looked at both Alphabet's press release page and Shutterstock's; done a Google search and looked on Yahoo Finance. I cannot find anything recent about a Google partnership (there were earlier deals in 2016 and 2021). The only detail comes from the transcript:

"In the second quarter, we signed a strategic partnership with Google that includes a five-year content and data partnership across Google products and services. Google selected Shutterstock for our global coverage of digital media to enhance product, and user experiences across Google."

Getting back to revenue. Their Q2 revenue was $111.903m from E-commerce, $79.637m from Enterprise and $17.3m from data. In other words the E-commerce segment (stock licensing via the web site) is their primary business, at least for now. Hennessy was careful to sound optimistic about future AI-fueled growth, once the photo-realistic content was good enough "...we strongly believe Shutterstock is positioned to be a winner in this new content type"

As you said, lots of chat about Giphy, but no detail about how SS could "...monetize [the] massive mobile audience base". I don't doubt the size of the audience, but his blather about "... moment marketing with real time conversations..." as a future growth area seems very vague and more wishing than planning. In any event I don't think contributors would benefit from this business at all unless content is used to create GIFs

Hennessy said that "...Shutterstock [is] benefiting from a dramatic uptick in customer demand and accelerating data supply from our contributor community." I watched a YouTube video this afternoon explaining how to upload AI generated content to SS even though it's not allowed and get it accepted - something the person claimed (he showed screen shots) to have done. Not sure how good it is if your increase in supply by contributors is of stuff you've explicitly said you don't want.

Later on Hennessy says " We went from 2.4 million in Q1 to 2.7 million contributors in Q2. This is the largest uptick we have ever had in contributors in the 20-plus year history of Shutterstock." I think that must be a flood of new contributors with AI stars in their eyes - where else would there be that many new illustrators, photographers, videographers wanting to sign up?

Apparently data licensing customers are coming back for more "Customers are also consuming refreshes of our training data, and there is an opportunity to provide them ongoing Metadata enrichment." Hennessy says "Looking ahead, we have a robust pipeline of data partnerships for the second half, also we have won over $85 million in bookings in calendar 2023". I assume the $17.3 million they declared for Q2 is part of that $85 million?

The CFO mentioned the softness in the E-commerce segment "While our E-commerce channel has been softer than expected, our rapid growth in data partnerships is more than making up for it..." From a contributor perspective, I think the problem is that they make much more money from licensing content than from the contributor fund. And the CFO also said "Our guidance assumes no recovery in E-commerce revenue growth rates this year"

As far as the data revenue, he explicitly said that the Q2 data numbers did not include any Google revenue, so that will come in the second half of the year

One of the analysts asked about future revenue from data licensing - if you make money this year, what does that mean if anything for recurring revenue in 2024 and beyond. The answer was long, circuitous and pretty vague:

"...our ability to grow this business year-on-year is really going to be a function of our ability to, one, expand our existing relationships and, two, bring new clients into the pipeline."

The investors want to know where the growth will be coming from and appear not to be sure if that can be data licensing if stock content isn't it any more.

https://investor.shutterstock.com/news-releases/news-release-details/shutterstock-reports-second-quarter-2023-financial-results

I haven't found a freely-available transcript of this morning's earnings call, but I'll update this if I do.

This article gives an overview of why investors weren't happy with the results:

https://stockstory.org/nyse/sstk/earnings/shutterstock-nysesstk-misses-q2-revenue-estimates

"Stock photography and footage provider Shutterstock (NYSE:SSTK) missed analysts' expectations in Q2 FY2023, with revenue flat year on year at $208.8 million"

"Shutterstock's revenue growth over the last three years has been unremarkable, averaging 9.53% annually. This quarter, Shutterstock reported rather lacklustre 0.95% year-on-year revenue growth, missing analysts' expectations."

" In addition, the company's gross margin also deteriorated."

Bear in mind that the crowing about subscriber growth omits mentioning what's noted in the detailed statements: Pond5 numbers were included for the first time in Q2 2023. I also think there's been conversion of existing customers from credit packs to subscriptions, which means it can be counted as subscriber growth, but it isn't really customer growth.

See all the charts in the appendix at the bottom of this investor page

https://content.shutterstock.com/investor-report/index.html

One number that stood out to me was the number of paid downloads had dropped 38.5 million, down from 42.7 million in Q1 2023 and 43.4 million in Q2 2022. Enterprise revenue was up, but E-commerce revenue was down - 12% lower than the same quarter in 2022.

"The decline in E-commerce revenue was primarily driven by continued weakness in new customer acquisition"

Enterprise revenue is still smaller than E-commerce and includes "...$17.3 million and $2.2 million, related to our computer vision data partnerships, for the three months ended June 30, 2023 and 2022..." Not sure what those partnerships are, but if they aren't royalty bearing activities, then contributors don't benefit from that type of growth.

"The increase in Enterprise revenues was primarily driven by growth in our computer vision data partnerships which generated $17.3 million during the second quarter..."

So I take that to mean that income from its stock licensing business was essentially flat and paid downloads were declining - hence the stock price dropping, I guess.

The problem is not new with stock photos. ...Unless it is editorial, people always have to be careful.

It's true that bad keywords are completely ignored by agencies and that  inaccurate descriptions have resulted in trouble for buyers. More often though it's careless staff not paying attention to accurate descriptions and using content based solely on looks. At its very worst, it was at least a photo of somewhere on planet earth, even if the keywords said Bahamas, Aruba, Guadeloupe, Jamaica, St Thomas...

Comparing problems with photographs to the dumpster fire of photo-realistic (ish) genAI is like comparing a gentle stream to a raging flash flood - the difference in degree is so great it's really a difference in kind.

In six months the genAI collection has ballooned and is full of things that aren't what they claim to be.

The image has been taken down now.

Yesterday there was even one more picture added, but this morning, I'm getting 404 errors if I click on anything in this guy's portfolio, so I think it's imploding.

And I thought it was totally out of line that they continued acceptances of new work after the reports of theft. It went from 905 to 999 to 1,103 to 1,104

Here's what I'm seeing this morning

I know that there will always be rejections and that contributors won't always agree with them, however with all the tales of rejects, I find a totally different story looking at recent genAI acceptances.

I would give a batch reject to all 5 images of a child holding out an empty bowl as not one image has a total of 10 digits. One could be explained by how the bowl is being held - maybe. The portfolio is small and new:

https://stock.adobe.com/contributor/211451045/lordrul

There are also two iMacs with logos in the portfolio (not in this morning's acceptances) which should have been rejected.

If you look at the other (not hand-related) details of several of those child-with-bowl images, they're a mess - focus in very odd places, stray lines above a tee shirt edge, lots of reasons to reject.