Managing Visual Media Risk In A GenAI World

Forbes

23-06-2025

Tal Lev-Ami is cofounder and CTO of image and video platform Cloudinary, which is trusted by more than 10,000 brands and 2 million users.
Over the past two decades, there has been a shift in how people prefer to consume information online, with visual content—images, video and interactive media—having gradually overtaken text-heavy formats. Research from Intergralads found that user-generated video alone is forecast to account for 82% of all consumer internet traffic in 2025.
As engaging and helpful as all this visual content is for consumers, it creates lots of technical challenges for those managing and delivering it. Brands today need to produce, retouch, share, moderate, publish and otherwise manage huge quantities of visual imagery across more channels than ever, making what was once a manageable nice-to-have a tech-powered, business-critical endeavor.
Inevitably, brands are relying on AI to automate the repetitive yet critical tasks required to ensure visual content looks great and performs well across every device. GenAI, in particular, is proving invaluable for a range of tasks like scaling content creation and personalizing customer experiences. Doing so brings enormous benefits, but it also poses potential risks.
According to Gartner's 2024 Hype Cycle for Emerging Technologies, GenAI has moved beyond what it calls the "Peak of Inflated Expectations." That's not to say we shouldn't still be inspired and ambitious about applying the technology; quite the opposite. It means we're entering a more practical phase, where the reality of execution means having to contend with the less shiny aspects of GenAI like governance, compliance and risk management.
After more than a decade into our journey with AI technologies, we've learned a lot about evaluating risk. Without question, companies that manage visual media at scale with GenAI are particularly exposed to errors, biases and compliance issues and need to have risk frameworks in place.
The simple framework below is designed to help evaluate the risk of using GenAI to transform a specific image and video asset or a collection of them.
• Scope: How many people are likely to see the asset? How many systems will be impacted?
• Centrality: This refers to the centrality of the asset to the user experience. Does it appear above or below the fold? Is this the most prominent/large asset or "eye candy" at the edge or corner?
• Probability Versus Severity Of Failure: Does the asset look a little bad or is there a risk it could harm the brand? For example, say somebody resizes the image and it distorts the aspect ratio—that's a little bad. But if the AI image is labeled in a racist or inappropriate way, that harms the brand.
• Fidelity Versus Appeal: Fidelity focuses on accurately representing the original product. It's especially important when customers need to assess details like texture, color, fit or quality—common in categories like fashion and apparel—where accurate images help reduce returns.
Appeal, on the other hand, emphasizes enhancing images to make them more eye-catching or emotionally engaging. This is often prioritized in social media, lifestyle branding and creative campaigns where the goal is to attract attention and drive engagement.
Those last two trade-offs are especially tricky to optimize at scale without putting automation and clear rules in place for humans and AI to follow, in line with risk tolerance. Every type and size of business will have different rules. News organizations, for example, prioritize visual asset fidelity over appeal.
By contrast, a luxury brand would likely prioritize appeal. A large enterprise that faces high scrutiny and regulatory risks would need strict human oversight of any AI automation. A smaller, more maverick business might decide to take more risks and "apologize later."
To manage GenAI risks at scale, clear policies must be established that define where and how your organization will (and won't) use AI. It's equally important to establish the role of human oversight to mitigate risk and meet brand and regulatory guidelines.
For instance, GenAI can assist in red-flagging obviously harmful user-generated content while escalating more nuanced decisions to human moderators. Similarly, GenAI can accelerate UI/UX workflows by generating initial layout concepts, which designers can then refine to meet aesthetic and usability standards. These should be reviewed on a regular basis (quarterly, for example) as things change very fast in the AI world.
AI models that perform well in controlled demos may falter when exposed to real-world complexity and high-volume traffic. That's why before moving to production, companies should rigorously test GenAI systems in their own environments, using real data and scenarios to uncover hidden weaknesses.
Incorporating A/B testing and shadow deployments can help identify unintended consequences early, before they impact users or brand trust.
Using AI provenance tools like watermarking and metadata tracking can help maintain accountability. To manage compliance, it's important to stay ahead of evolving regulations on AI-generated content.
Optimizing the workflow with GenAI is essential for harnessing visual media's many benefits, but doing this at scale safely requires proactive risk management. By aligning AI adoption with structured risk frameworks, companies can innovate at hyperscale while maintaining trust and quality.
Businesses that scale AI responsibly, treating risk assessment as a core strategy rather than an afterthought, definitely stand to gain the best return on their visual media spend.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?