What Getty's Prepackaged Chapter 11 Says About Selling Images

On 12 October, Getty Images announced a restructuring agreement with all of its major creditor groups. The Getty family and Koch Equity Development will put up to $533 million of new equity into the business. Holders of 83.5% of the unsecured bonds agreed to retire $270 million of senior unsecured notes. Secured lenders representing roughly 89% of that debt, and every revolving credit lender, accepted lower interest and later maturities. When the process finishes, the two investors own 100% of the company and the existing shares are worth nothing.

As of this writing the cases have not commenced. Getty's own milestones require it to begin soliciting votes and to start the Chapter 11 cases no later than 21 October, with the restructuring effective within 75 calendar days of that commencement. If you are reading this after that deadline, check the bankruptcy docket rather than the press release. The filing is prepackaged, which is the corporate equivalent of arriving at the negotiating table with the signatures already collected.

Spanish coverage from Xataka called it the first large company of the past to fall because of AI. That framing is worth pushing on, because the mechanics tell a more specific story than a fall. A prepackaged Chapter 11 with creditor consent is a recapitalisation run by the lenders, not a liquidation, and Getty says the point of the exercise is to clear legal liabilities while operations continue. Contributors, vendors, customers and staff are, per the company, unaffected. So what actually gave way is narrower than the headline: a business model stopped paying for itself while the assets and the workflow kept running.

The numbers that explain the filing

Getty's own financial update, as reported, put annual subscribers at 241,000 in June 2026, down from 321,000 a year earlier. That is a quarter of the annual subscriber base gone in twelve months. The shares had already been suspended from the NYSE in late September over an abnormally low price, and moved to the OTC market on 30 September. By the time the deal was announced, the market value was in the tens of millions, from the billions the company was valued at when it re-listed in 2022.

Then there is the merger it did not get. Getty walked away from a $3.7 billion acquisition of Shutterstock after the UK competition authority required Shutterstock to divest its editorial business, a condition Getty called unacceptable. The company reported spending roughly $105 million on the failed acquisition and about $17 million in ongoing costs from its copyright case against Stability AI. The restructuring is explicitly designed to dispatch legal liabilities, which is a polite way of saying that the lawsuits and the debt structure, not the photo library, were the weight that needed removing.

What a stock library actually sells

Most explanations of Getty's decline stop at "AI generates images now." True, and too shallow to be useful. A stock library sells a licence to use one specific file, and buyers came to it because the alternatives were worse: commission a photographer, hire an illustrator, or shoot it yourself. Generative models did not delete stock photography. They removed the scarcity from the middle of that list. The same pressure had already reached performers, which is how actors started losing roles to their own AI clones before anyone worried about the picture library.

You can see the shift in what Getty started advertising. Its generator is trained on licensed content that carries indemnifications and perpetual worldwide rights. Read that sentence back: the product being sold is no longer the picture, it is permission. Indemnity, provenance and a rights posture a legal department will accept are things a generated file from an unaccountable pipeline cannot offer. That is a real product, and it is a service business wearing an image business's clothes.

The legal route went worse than the licensing route, and it is still running. Getty opened proceedings against Stability AI in the English High Court in January 2023, claiming that more than twelve million of its photographs had been scraped to train Stable Diffusion, and filed a parallel case in Delaware the following month. In the UK trial it dropped its primary training-infringement and database-right claims partway through, and in November 2025 the High Court rejected the secondary copyright claim, leaving a limited trademark finding over outputs that reproduced a distorted Getty watermark. Getty has permission to appeal. In the United States it voluntarily dismissed the Delaware action in August 2025 and refiled in the Northern District of California, where the court largely denied Stability's motion to dismiss in April 2026, fact discovery runs into late 2026, and trial is scheduled for January 2028.

So the picture is not a company that sued and lost. It is a company whose enforcement went slowly and partially, which is a different problem when the licensing side of the business is eroding quarterly. Alongside the litigation, Getty has signed a multi-year arrangement that puts its archive in front of Perplexity answers and an OpenAI deal announced in June whose public terms appear to exclude model training. Distribution through the products that displaced the library has moved faster than the courts have.

Consider the watermark, the most visible part of the old model. It was never a control. It was a reminder aimed at buyers who intended to pay. Removal tools are ordinary, and plenty of public figures happily repost their own licensed frames with the watermark intact, because their interest is in the picture, not the licence attached to it. Using the watermark as a moat means betting that nobody on the internet is willing to spend thirty seconds.

The contrast inside the same industry is instructive. Freepik, the Malaga-based graphics search engine, did not sue anyone. It bought Magnific, a generative upscaling and editing startup, in 2024, renamed the whole company after it in April this year, and rebuilt itself as a marketplace for other companies' models. It now competes with Adobe rather than with the idea of generated imagery. Both companies read the same market. One of them bet on being right about the past.

Three things worth taking from it

If you sell stills or footage, the per-file licence is the piece of the business that stopped working, and it is not going to start again. Price the outcome and the labour: a specific shoot, a specific retouch pass, a specific deliverable for a client who needs it on Thursday. Files are nutritionally empty as a product line once the buyer can produce a near-enough version in an afternoon.

Commercially safe is a feature, and it is more sellable than style. A large client will pay for a documented chain: where the source came from, what rights you hold, which model version produced the file, and when. Keep that record as you work, because you cannot reconstruct it in a meeting.

And watch where the money ended up. It is not with the party that owned the archive or the party that won the argument about training. It is with the parties that own distribution. That is a durable observation about this market, and it does not require a bankruptcy filing to be true.