The Risks and Rewards of Generative AI for Visual Media
Online consumers have an insatiable appetite for images. Yet, until now, that content has been expensive and time-consuming to create and deploy across channels. The emergence of Generative AI changes all that. It allows brands of all sizes to finally create content at the velocity and scale that the modern economy demands. It opens up new use cases beyond original asset creation that will likely deliver greater immediate value. It also introduces new complexities, challenges, and risks. So, what are some ways to get started with Generative AI for visual media and how can brands take advantage of the rewards while mitigating the risks?
Use Cases for Generative AI Images
Brands have an opportunity to use Generative AI for images in a few ways. The first is obvious: they can use it to create whole pieces of content from scratch. But they can also use it to create, enhance, and sharpen background environments of existing content to serve a wide variety of different uses. While each of these use cases have both benefits and challenges, there are significantly more risks and challenges associated with the “from-scratch” creation.
Generate Full Images - Risky and Complicated
Creating images from scratch offers the advantage of creating ample amounts of content at record speed. This normally happens through a text-to-image creation process: users supply a textual description of the required image (prompt) and the AI uses that context to create a “complete” image.
Given businesses spend a great deal of time and resources creating visual content, it’s easy to see the benefits Generative AI can bring for from-scratch images creation.
But this also carries a very high degree of risk. When creating an image, Generative AI does its best to imagine the entire scenario, but it might not imagine the core product correctly, which is detrimental to the success of the campaign.
Even worse, it also cannot reliably cross-reference the image against brand guidelines, quality standards and composition rules, which introduces another layer of risk. A fully generated image can be good for use as a background but it is rarely the right way to present the main product.
Accelerate Customization - Safe and Scalable
Using Generative AI for the background composition of images is a much less risky use case—and a much more accessible one for most brands. In this use case, brands can take an existing image and use Generative AI to enhance it.
For example, it can be used to:
- Generating new content that doesn’t yet exist beyond what’s already in frame. This is referred to as outpainting. Think of editing a picture taken on a phone in portrait mode that needs to be viewed horizontally to fit a screen layout. The traditional answer was to crop based on the centerpoint, reduce the resolution, and risk cutting out some of the important parts. With generative AI, brands can generate new, contextually consistent content to seamlessly fill in the space not captured in the original asset. Here is an example of outpainting before and after.
- Paint a new background, which can be particularly useful if a product was shot in a subpar environment that needs to be amended, or to add color or detail to a pure white background.
- Remove and replace unwanted pixels with new pixels that compose a more compelling or relevant scene - a process called inpainting. Here is an example of inpainting before and after.
These are all inherently much less risky than using Generative AI to create the entire image from scratch because they keep the integrity of the core product/subject and use the tool to fill in just the secondary elements.
Unfortunately, AI simply cannot be trusted to create an entire product image based off of a text description at this point. But, brands can use AI to take that product image and place it in different scenes by generating new backgrounds. That way, the brand can keep the exact representation of the object in focus while easily creating multiple versions of the background that would be hard, or even impossible, to create otherwise.
Generative AI Still Requires A Human Touch
It’s important to remember that no matter how brands choose to use Generative AI for images, creating unique, on-brand visual assets still requires humans to manually review, edit, and optimize each asset after creation. Assets need to be compressed, cropped, resized, and formatted to make them perfect for each and every user’s viewing context. This limits the velocity gains of using generative AI, unless other AI-enabled automation tools are included as part of the workflow.
Brands should look to incorporate optimization tools into the generative AI workflow which will help free up developers’ time and automate tasks that would otherwise use up their valuable time and bog them down.
AI is Part of a Visual-First Future
Artificial intelligence is already integrated into the creative workflow for most of the world’s top brands - from organization to moderation to optimization and delivery. But the influence of generative AI is still quite small when it comes to its impact on visual media. It’s exciting to see use cases such as outpainting and inpainting emerge as valuable tools, and the future is certainly bright for this nascent technology.
As consumer demand for incredible visual experiences continues to rise, AI tools - including generative AI - will continue to evolve and become even more critical to success. Just don’t expect generative AI to revolutionize the creative process overnight.
[Editor's note: This is a contributed article from Cloudinary. Streaming Media accepts vendor bylines based solely on their value to our readers.]
Related Articles
Media and entertainment companies of all sizes struggle to attract and retain audiences eager to find and consume content that interests them. Because content is abundant with too many options to navigate and pay for—and switching costs are low—profitability has eluded many providers. How can media and entertainment companies respond? Many are turning to a new ally: AI.
29 Jan 2024
Leon Huang, Senior Director of Sports, Media, and Entertainment at SAP, discusses the ethical and legal implications of AI for the entertainment industry addressed in the sinister Netflix Black Mirror episode "Joan is Awful," and what steps can be taken to establish ethical standards and transparency around the use of AI in media.
30 Aug 2023
Josh Dorward of Cloudinary discusses the ways that AI is helping marketers embrace top video trends.
06 Dec 2022
With Disney and Netflix becoming the latest providers to succumb to the inevitable commercial realities of the subscription-only TV business model, what does this mean for both customer experience and advertising effectiveness?
18 Nov 2022
From content discovery to video indexing, AI can enable customization and hyper-personalization of the OTT viewer experience and help OTT programmers and marketers learn more about their audience.
25 Jul 2022
Manual analysis of live streaming content to identify key highlights is both time-consuming and involves expensive manpower. For a growing number of content developers, the answer is bite-sized videos created with the help of AI.
11 Jan 2022