Getty Images and OpenAI have reached a display partnership, authorizing stock image content to be integrated into ChatGPT’s search and discovery scenarios. This is not a training license, but rather inserting licensed images into generated results—after three years of copyright battles, the two sides have finally sat down to do business.

OpenAI Buys Images from Getty: ChatGPT Finally Won’t Have to "Imagine" Pictures Anymore

On June 21, Getty Images (NYSE: GETY) officially announced that it had signed a display licensing agreement with OpenAI. Simply put, from now on, when you ask ChatGPT “Show me a photo of the opening ceremony of the Paris 2024 Olympics,” what pops up will no longer be some blurry, AI-fabricated pixels, but the actual photo taken by a Getty photographer on site—complete with attribution, metadata, and a link back to Getty’s website.

At first glance, this is just a conventional content licensing deal, but considering the past three years of disputes between AI companies and stock photo copyright holders, the signal is significant.

This Is a "Display" Business, Not a "Training" Business

Let’s clarify the concept. Content licensing in the AI industry broadly falls into two categories:

Training license: You pay, and I feed the text/image/video into the model as training data. For example, Shutterstock’s six-year renewal agreement with OpenAI in 2023 was of this type—OpenAI used images to train models, Shutterstock collected licensing fees and shared revenue with the creators whose works were included.
Display license: You pay, and I present the content in your product interface, with attribution and backlinks. Essentially, it’s closer to how search engines display image results, without touching training data.

Getty’s agreement this time is the latter. Getty Images CEO Craig Peters has repeatedly emphasized over the years that the company does not sell training data. In early 2023, Getty even sued Stability AI in both London and Delaware—accusing Stable Diffusion of scraping 12 million photos from its collection without authorization. At the same time, Getty partnered with NVIDIA to develop Generative AI by Getty Images, using “fully licensed corpus for training, and legally protected generated images” as their approach.

So this collaboration with OpenAI is essentially Getty treating ChatGPT as a new distribution channel. You could think of Getty setting up a “booth” inside ChatGPT, and OpenAI acquiring a batch of high-quality visual content to enhance its search and discovery experience—without either side touching the other’s "underwear": Getty’s training data remains off-limits, and OpenAI’s model weights remain unchanged.

Why Does OpenAI Need Licensed Images Now?

The answer is counterintuitive: not because GPT image generation can’t do it, but precisely because it can—and that causes problems.

ChatGPT’s search functionality has been live for over a year, competing with Google and Perplexity for real-time information retrieval. Text retrieval, with crawling + RAG + model integration, has caught up in terms of experience. But for visual information, previously ChatGPT used two methods:

Scraping images from public webpages: copyright status unclear, common cases of watermarks, low resolution, or incorrect images;
Generating directly with GPT-Image-1: You ask “What does a photo of Trump and Musk together at the White House look like?” and the model draws one. Generating “factual images” about news, people, or places is high-risk—fabricating a presidential photo carries far greater consequences than text hallucinations.

After integrating Getty, both pitfalls can be filled. Queries about news, celebrities, sports events, or historical archives can directly pull licensed Getty images. Getty has 500 million photos and tens of millions of videos, covering everything from Pulitzer-level photojournalism to commercial stock. In the public market, only Adobe Stock and Shutterstock rival Getty in scale.

It’s worth noting: OpenAI choosing Getty instead of Shutterstock for display licensing may be related to positioning. Shutterstock leans toward UGC and commercial assets, while Getty has deeper resources in editorial and news photography—the exact segment ChatGPT’s search lacks most.

The Numbers Each Side Is Running

These deals often look mutually beneficial, but behind the scenes, each side’s calculations are different.

OpenAI’s goals:

Fill the visual gap in search experiences to compete with Google SGE and Perplexity;
Significantly reduce copyright risk. Getty’s metadata-equipped images in ChatGPT act as a legal safeguard;
Build a sustainable content ecosystem for “AI search”: You license content to me, I send traffic to you—similar to the Google News model.

Getty’s goals:

Gain a new traffic channel. Traditional stock photo B2B clients (media, advertising, design firms) have been heavily eroded by generative AI in recent years—finding new distribution scenarios is essential;
Add funds to the royalty pool for photographers. Getty has long shared revenue proportionally with photographers; display fees from ChatGPT will in theory go into the pool;
While still litigating against Stability AI, establish a “legitimate partnership” as a flag—showing “I’m not against AI, I’m against unpaid AI.” This strengthens their courtroom narrative.

Industry Map Perspective

Over the past two years, OpenAI has been “shopping” for content partnerships, with Getty just the latest SKU:

| Date | Partner | Type | Purpose | |------|---------|------|---------| | 2023.07 | Shutterstock | Training + Display renewal | Images/videos/music to train models | | 2023.07 | Associated Press | Training | Text corpus | | 2024.04 | Financial Times | Training + Display | Model training + ChatGPT news summaries | | 2024.05 | News Corp | Training + Display | $250M over five years | | 2024.05 | Vox Media / The Atlantic | Training + Display | Content ecosystem | | 2024.10 | Hearst | Training + Display | Fashion and business content | | 2026.06 | Getty Images | Display | ChatGPT search images |

See the pattern? OpenAI has swept through top publishing groups on the news text side, leaving only those like The New York Times (still in court) unsigned. On the visual side, things have been slower—Shutterstock’s training license was signed in the Dall-E era, better suited for generation. This Getty deal fills the “search display” slot, aligning another piece of OpenAI’s copyright puzzle.

A Detail: What Does This Mean for Generated Images?

Worth noting: This agreement is strictly for search and discovery scenarios, not for training image generation models. Meaning, when you generate with GPT-Image or Sora, the model weights do not contain Getty’s corpus.

This boundary is drawn very clearly, underscoring Getty’s wariness toward training. Their Generative AI by Getty partnership with NVIDIA is still running, where Getty controls the training data and generation rules. Selling images to OpenAI for search results is one thing; feeding them to GPT-Image is another—Getty likely won’t approve the latter.

This “sliced licensing” method may become the standard posture for future negotiations between visual copyright holders and AI companies—splitting licensing by scenario: display for one price, training for another, derivatives for yet another. The era of “bundled sale” is ending.

Developer Perspective: What Can You Use?

Developers integrating the ChatGPT API are unlikely to feel direct effects from this deal in the short term—display licensing is currently restricted to search and discovery within the ChatGPT product, with no new API endpoints released. If you’re building RAG or search-enhanced apps and need licensed image sources, Getty’s own API remains a commercial option; if you want free sources, Unsplash and Pexels (CC0) are still preferred.

However, one potential impact is worth watching: as ChatGPT search’s image quality improves, downstream apps that directly relay ChatGPT’s search results could face compliance issues with redistributing Getty’s content. OpenAI will likely follow up with terms.

Side note: OpenAI Hub (openai-hub.com) has always supported ChatGPT’s backend GPT-4o and GPT-5 series models, with direct domestic access and full OpenAI-format compatibility. Search calls run via chat.completions with tools. After Getty’s integration, image display in the ChatGPT product will update, but API call logic remains unchanged.

Final Thoughts

The Getty–OpenAI deal’s financial terms weren’t disclosed, and its scale may not match News Corp’s $250 million. But its significance lies in: The first time a visual copyright holder officially entered ChatGPT’s product as a “display partner”, rather than as a “training data provider” or “plaintiff.”

The pervasive sentiment in the visual creator community over the past three years—that “AI is here to steal our jobs”—may loosen slightly from this deal. Not that conflicts are resolved, but that people realize: arguments aside, business still has to be done, and rules still need to be negotiated.

Stability AI’s lawsuit remains undecided, while OpenAI has already signed the deal. That’s the reality of commercial timing.

References

ITHome: Getty Images and OpenAI Collaborate to License Stock Content into ChatGPT – Full domestic media coverage of the cooperation news, including background on Getty’s earlier collaboration with NVIDIA and OpenAI’s historical content deals.