Anyone who creates digital media for publication understands the challenges of generating photos, videos, illustrations, and vector graphics from scratch. Doing so gives you creative freedom and infinite customizability, but there’s often a significant trade off: it takes a lot of time and money to create all the elements of a creative project yourself. We live in a world of project deadlines, and we often don’t have the luxury of time; a lot of us are also resource constrained, and orchestrating your own photoshoot may be well out of budgetary reach.
This is why stock photography and video have become so popular; they provide a necessary resource for content creators who need to source assets quickly and affordably. However, the advantages that come with having a large content repository can also lead to new challenges. Sorting through all those assets to find the one that’s just right can be a formidable and frustrating task. Moreover, the time spent on this task is time that could be better spent on the rest of the content creation process and strategy.
In fact, according to a qualitative research study commissioned by Adobe in October 2018, based on interviews with creative pros across the US, UK and Germany, 74% of creatives said they spend over half of their time on repetitive, primarily uncreative tasks.
That’s why at Adobe, we leverage Adobe Sensei, our artificial intelligence and machine learning technology, to take the pain out of mundane tasks like this in Adobe Stock. Stock has a digital content library containing more than 150 million assets, it gives you the ability to access images, videos, templates, and even 3D objects to find the ingredients you need to create, without having to produce content from scratch. We want to ensure that content search and discovery will always be relevant and efficient, so that you’ll get to spend more time doing what you do best: creating.
Solving for the time-consuming process of traditional image search
Through interviews with Adobe Stock customers, we’ve learned that the search process is one of the biggest pain points when digging into deep content libraries to find the right assets. Project owners and their requirements are often quite particular, and as a result this process can take upwards of 30 minutes per asset. This work is typically done with highly trained and well paid workers, and the search experience should be more efficient and much more pleasurable for them.
Many opportunities for improvement trace back to the use of conventional text-based search, for images. Under the hood, these text search queries are mapped to human-assigned text metadata that are associated with each potential result. While this type of search architecture is successful for helping us surface content which is generally relevant to the text query, images often have unique features that are hard to capture through written descriptions. For example, sometimes it’s hard to articulate what exactly it is that attracts you to that perfect image, but when you see it, you instantly know it’s what you’ve been searching for. In addition, for many of the image attributes that designers care about, there is no standard search vocabulary. ‘Depth of field,’ ‘bokeh,’ and ‘background blur’ are all ways that a designer could potentially search for a depth of field effect. Similarly, ‘copy space,’ ‘white space’ and ‘negative space’ are all different terms that a designer could search for images with space to overlay text. This too, makes it hard for traditional search engines, especially since the lack of consistency in the search terms also exists across the metadata.
We want browsing Adobe Stock to be relevant and meaningful at all times. This is why we have created a series of Adobe Sensei services to improve search functionality and cut down on wasted time and resources. These services, powered by artificial intelligence, are trained to “see” an image in the complex way a customer does, by discerning all of its relevant visual characteristics. Below is a breakdown of four features built off of services the Sensei & Search team has developed in collaboration with the Adobe Stock and Adobe Research teams. By using these features, content creators can identify the image that perfectly suits their needs much faster than before.
Visual Search — Find Similar Controls
Visual Search in Adobe enables a customer to use an image they have or find in search results to generate a set of search results that are similar to that image. The Find Similar Controls feature adds an extra layer of functionality to this: you’re able to pick elements of an image, like the color or composition, and find images that have that same visual characteristics. This also allows you to tweak your visual search; e.g., if you want a windsurfing image with the windsurfing board on the left but a slightly different aesthetic than the original image you found, you can do so with the click of a button.
