What occurs when an organization desiring wealthy metadata for its multimodal AI fashions joins forces with an international chief in ingenious content material? The solution lies within the strategic partnership between Reka and Shutterstock.
On this publish, we’ll discover:
Shutterstock Companions with Reka
On this planet of tech, the AI panorama is the recent position to be at this time. Then again, as new corporations try to outpace their competition hastily, once in a while a little bit out of doors lend a hand can reap rewards.
Reka confronted a problem: Its group had to educate its multimodal AI fashions however required get entry to to huge quantities of information. High quality metadata performs a a very powerful position in coaching by means of bettering the style’s talent to grasp, interpret, and generate content material throughout other modalities, equivalent to textual content, photographs, and video.
Licensing Shutterstock information would allow Reka to make stronger the accuracy of its information labeling, spice up the potency of its style coaching, and facilitate complicated options like personalization and customization of its AI.
On the similar time, Shutterstock was once in search of an effective and efficient option to strengthen the metadata tied to masses of thousands and thousands of its library belongings. Doing this might make it more uncomplicated for its consumers to discover a particular symbol or video at the Shutterstock market. Wealthy and numerous metadata would additionally result in higher categorization and visibility of a contributor’s merchandise at the market, doubtlessly riding extra gross sales, maximizing the asset’s income attainable, and bettering product discoverability for everybody similarly.
So, in June 2024, Shutterstock shaped a multi-year partnership with Reka to license information from its huge library of belongings to increase Reka’s frontier-class multimodal language AI fashions. In go back, the AI corporate would strengthen the metadata of Shutterstock’s 550 million belongings throughout photographs and video.
Reka Multimodal AI
When you’re unfamiliar with multimodal AI, those fashions can analyze a couple of kinds of inputs concurrently. In easy phrases, you’ll be able to supply them with photographs, movies, audio, PDFs, and textual content, and they are going to generate extra advanced and nuanced outputs.
Right here’s an instance of the way Reka’s Core style works in observe:


Then again, to obtain a correct solution, the AI fashions require a huge quantity of information, and it’s no longer simple to obtain ethically sourced, blank, and enriched information for coaching multimodal AI fashions.
When you stay alongside of information about synthetic intelligence, you’ve almost definitely heard about tech corporations “scraping the internet” for information. But even so angering content material creators, scraping public information too can result in attainable proceedings that can derail even essentially the most tough fashions.
That’s why increasingly corporations are choosing legally approved information that they may be able to depend on, keeping off attainable proceedings for violating website online phrases of carrier, copyrights, and privateness insurance policies.
Complying with all appropriate regulations can also be a large number of paintings. Reka known Shutterstock as an international chief in visible information, providing the most efficient dataset for coaching multimodal AI fashions, which streamlined the advanced procedure for them.
Moral concerns abound on this house, together with information dealing with, consumer coverage, bias prevention, and making sure moral practices at each level of information control.
“By means of the use of ethically sourced information from our library of greater than 670 million belongings, we’re serving to Reka succeed in its function of advancing the AI building of its fashions and, in go back, we will be able to additional make stronger our library. Making improvements to the metadata to our asset library will perform a little good stuff for patrons and participants, equivalent to expanding the probabilities of belongings being listed and ranked by means of search engines like google, each in our market and externally,” says Shutterstock’s Senior Director of Product Jergan Callebaut.
Shutterstock expects this multi-year partnership to proceed to blossom as we glance to the long run. Greater than 60 million new belongings are added to the Shutterstock library yearly, all of which is able to want well-tagged and described information. With Reka’s lend a hand, consumers will spend even much less time looking and extra time developing.
Take, for instance, the Shutterstock symbol under.

The picture description is “Lady pouring honey onto skinny pancakes with berries at desk.”
However with Reka’s AI including extra data to the picture metadata, it may possibly even have such things as:
- Environment: An indoor atmosphere, most probably a kitchen or eating house, with a focal point on a meal preparation or serving scene.
- Visible taste: The visible aesthetic is heat and alluring, with cushy, herbal lighting fixtures suggesting a homey environment.
- Colour palette: Brown, beige, yellow, inexperienced, pink.
- Items: Stack of crepes, honey, blueberries, cherry, mint leaf, wood bowl, honey dipper, plate.
- Spatial relationships: The stack of crepes is targeted at the plate, with honey being poured over it. Blueberries and cherries are positioned across the crepes, and a mint leaf is on most sensible. Within the background, a wood bowl and honey dipper are visual.
- Detailed description: This symbol captures a second of meal preparation or serving, that includes a stack of crepes on a plate with honey being poured over them. Surrounding the crepes are recent blueberries and cherries, with a mint leaf including a pop of colour and most probably a recent taste. The wood bowl and honey dipper within the background counsel that the honey is freshly poured. The atmosphere seems to be a homey kitchen or eating house, with cushy, herbal lighting fixtures making a heat and alluring environment.
Shutterstock Knowledge Upgraded
Shutterstock information licensing has received momentum over the last couple of years and is now depended on by means of corporations equivalent to NVIDIA, OpenAI, Meta, and LG. Our information makes use of all kinds of content material varieties, together with 620M+ photographs, 50M+ movies, 3M+ song and audio tracks, and 1.2M+ 3-D fashions.
So, whether or not you want metadata to construct your generative AI fashions, licensable information for system studying, or need to educate content material moderation running techniques on Shutterstock’s numerous array of pictures, the standard of its best-in-class metadata will handiest recuperate.
Shutterstock could also be proceeding to spend money on its intensive library of human-created and reviewed metadata, however with Reka’s lend a hand, they’re additional detailing the metadata and making it extra nuanced for information services and products purchasers, which can result in upper click-through charges and boosted consumer engagement of their very own web pages and merchandise.
If you wish to get began, you’ll be able to touch Shutterstock immediately or paintings via Amazon Internet Services and products and Google Cloud. Its group is worked up that can assist you ideate, curate, and customise datasets in your wishes.
In reality, Reka applied Amazon Internet Services and products to obtain the information. “The S3 supply via AWS has been so clean. We admire how powerful, protected, extremely scalable, and dependable it’s,” says Eric Chen, Head of Implemented at Reka.

Impactful Knowledge
“This partnership is a testomony to the ability of collaboration. By means of operating with Shutterstock, no longer handiest are we streamlining the advanced technique of obtaining information for our AI fashions, we also are handing over higher fashions for our consumers,” says Dani Yogatama, CEO and Co-Founding father of Reka.
License this background quilt symbol by means of Shutterstock AI Generator.
This publish was once in the beginning revealed onSeptember 20, 2024
Lately seen
${excerpt}