The images on these articles come from a model running on my own laptop. I wrote down why a while back: no upload, no queue, no supplier changing the price while I am working.
Since 12 August the other route sits in the same window. Shutterstock has a connector in Claude, so you can pull licensed material without leaving the conversation. That sharpens the choice, because generating and buying now stand literally side by side.
What was announced
Shutterstock announced the connector on 12 August 2026. From inside Claude you search their library for images, video, music and sound effects, and license whatever you need. The motive is openly commercial: the company wants its rights-cleared material to become part of AI workflows, the same way it put an app inside ChatGPT earlier.
Two things stated in the announcement that you want before getting excited.
Searching is free, licensing is not. Shutterstock’s own page about these integrations says it plainly: searching the library from Claude costs nothing, but using an asset commercially takes a subscription or a purchase. The connector is a shop window, not a discounted till.
The terms of the deal are not public. MarketWatch reported that neither party disclosed what they agreed. Who pays whom, and whether Anthropic earns anything here, is unknown.
The two routes side by side
| Generate locally | Buy a license | |
|---|---|---|
| Provenance | a model on my machine | a person who made it |
| Rights | unclear and different per jurisdiction | settled by contract |
| Cost | electricity and a one-off install | subscription or single purchase |
| Dependency | none, it runs offline | account, subscription, connector |
| What you get | exactly what you describe, or something adjacent | what exists, photographed by someone |
| Who gets paid | nobody | the maker, a share |
That last row is the most interesting one, and also the one nobody leads with.
The number nobody opens with
“Rights-cleared” means the rights are settled. It says nothing about how much reaches the maker.
Shutterstock runs a tiered royalty for contributors: the more licenses you sell within a calendar year, the higher your percentage. According to an overview from June 2025, that ladder runs from 15 percent at the bottom to 40 percent at the top, and it resets to the lowest tier every 1 January. Since April 2025 there has also been a pilot around an unlimited plan where contributors share half of net revenue.
An important caveat: those percentages come from third-party analysis, not from a Shutterstock page. I could not confirm them in the official contributor terms, and they may have changed since 2025. Treat them as an order of magnitude, not a contract.
Even with that caveat it changes the question. The choice is not “free versus paid”. The choice is: pay nobody, or pay someone a share that at the bottom of the ladder is one sixth of the license price.
The same library, both sides of the table
The same press release describes a second arm of the business, and it deserves attention. Shutterstock does not only sell licenses to makers like me. It also sells training data to whoever builds AI models: rights-cleared, labelled, with what they call clear data provenance.
That is not an accusation, it is simply stated there. But it does mean the same library is sold in two directions. To me as images I may use, and to model builders as material for training systems that can approximate those same images. The contributor sitting in both streams gets a percentage per license in one, and something they rarely have visibility into in the other.
What I actually do
I am switching the connector on and I will use it, but not for the covers here.
For the covers nothing changes. They are illustration, not documentation: nothing on them needs to be real, and the local model produces exactly the risograph look this site is built on. A licensed photograph would fit worse and cost more.
Where the connector wins is anything that has to be a photographic fact. A building that genuinely looks like that. A product that genuinely looks like that. A situation that genuinely happened. A generator can suggest those things convincingly, and that is precisely the problem: convincing is not the same as true. Then a license on a photo somebody actually took is the better route, with the bonus that the rights are settled.
The rule I take from this: generate when the image illustrates something, license when the image proves something. That is the same boundary as in last month’s piece on watermarking. Provenance starts to matter the moment somebody relies on the image.
The caveat
The default shifts from choosing to grabbing. As long as buying an image lived in a separate tab, it was a decision: you went somewhere, you searched, you paid. In the same window as the generator it becomes a suggestion among two other suggestions. That makes the good choice easier and the thoughtless one easier too, and those are not the same motion.
Responsibility for the license stays with you. The connector makes searching easy; it does not determine which license you need. Standard or enhanced, merchandise or not, what print run: that distinction still exists and it does not appear in your conversation. Anyone assuming “rights-cleared” means “I can do anything with this” finds out when it goes wrong.
What drains away is sight of the maker. On a stock site you still saw a name, a portfolio, sometimes a country. In a conversation you get a result. The chain of who made this, what they earn from it and under which terms gets thinner with every layer in between. I use a local model where that chain is absent entirely, so I say this with a plank in my own eye: the honest comparison is not between clean and unclean, but between two different ways of losing sight.
Frequently asked questions
Does the connector cost money?
Searching does not. Shutterstock states that searching the library from Claude and ChatGPT is free. The moment you want commercial use you need a subscription or a single purchase, and which license you get depends on your plan.
Which plans does this work on?
Directory connectors work on Claude, Cowork, Desktop and mobile. On Team and Enterprise an administrator has to enable the connector for the organisation before anyone can link it; after that everyone authenticates individually.
Does this replace generating images yourself?
No, it puts something beside it. The choice depends on what the image has to do. If it illustrates an idea, generating is faster, cheaper and easier to steer toward a house style. If the image has to stand for something, you want what somebody actually recorded.
Am I sure I can use licensed material for my purpose?
Not on the strength of the connector alone. “Rights-cleared” says the rights are settled, not that every application is covered. Merchandise, large print runs and resale fall under different licenses than a blog post. That distinction is yours to make.
Sources
- Shutterstock, “Shutterstock Brings High-Quality Images, Videos, Music, and Sound Effects to Anthropic’s Claude”, press release, 12 August 2026 — investor.shutterstock.com
- Shutterstock, “LLM Integrations”, accessed 13 August 2026 — shutterstock.com/discover/llm-integrations
- Claude Help Center, “Use connectors to extend Claude’s capabilities”, accessed 13 August 2026 — support.claude.com
- MarketWatch, on the undisclosed terms, 12 August 2026 — marketwatch.com
Checked on 13 August 2026. I read the press release and Shutterstock’s integrations page myself; that searching is free while licensing is not appears there verbatim. What I could not verify is the contributor ladder: the 15 to 40 percent comes from third-party analysis dated June 2025, not from the official contributor terms, so treat it as an order of magnitude. The terms between Shutterstock and Anthropic were not disclosed. As far as I can tell Anthropic published no announcement of its own about this; everything comes from the Shutterstock side, which means the framing is the seller’s.