Generative AI image and video tools advanced rapidly this year, edging closer to the uncanny valley as platforms like OpenAI’s Sora and Google’s Nano Banana expanded across the market. Marketers now expect AI to streamline workflows, cut costs and plug directly into existing creative pipelines rather than function as novelty tools.
“If you’re using AI to automate a certain task, but then you have to move files from 15 different softwares, you’re not actually making your life more efficient,” said Freddy Dabaghi, chief transformation officer at Crispin.
To understand which platforms are delivering, Digiday compiled a 2025 agency generative AI report card, ranking leading tools on output quality, user experience and capabilities. The list is not exhaustive, as new products continue to enter the market.
Digiday contacted Google, OpenAI and Midjourney for comment. Google and OpenAI defended the creative potential of their respective tools, while Midjourney did not respond by press time.
Nano Banana: A
Marketers describe Google’s Nano Banana Pro, launched in November and built on its Gemini large language model, as a frontrunner in image generation and editing. At least two agency executives say it has become their default starting point for creative workflows, citing higher precision than Midjourney, Canva or Adobe Firefly.
Marketers report less noticeable “AI sheen” — inconsistencies or an overly polished look that signals AI involvement — and more reliable results. Images are often hyper-realistic, making it harder to distinguish AI creations from real photography.
Integration is another key advantage. Nano Banana is embedded across Google Ads, Gemini and other Google Workspace products, allowing teams to move more seamlessly from ideation to live campaigns.
“Nano Banana is a pretty dang good image generation model,” said an agency creative director who spoke on background, adding that Nano Banana is the agency’s “favorite and preferred image generation tool.” The agency is a large global digital marketing shop within a holding company.
Its limitations reflect broader gaps in generative AI rather than specific flaws with Nano Banana itself. Marketers still struggle to maintain strict consistency for storyboards, characters and narratives across multiple iterations; slight prompt changes can drastically alter AI-generated creative.
VEO: A
VEO, Google’s text-to-video model, also ranks at the top of marketers’ toolkits.
“VEO is probably the one that we use the most in that it just is incredibly robust,” said a second anonymous source, an executive at a branded entertainment company. “It’s just pretty damn good at everything.”
Initially released in late 2024, VEO gained momentum after audio features rolled out in VEO 3.1 in May. Marketers praise its ability to maintain character consistency, accurately interpret prompts and handle cinematographic concepts. New voice, audio and lip-sync features have further improved results, and some say it now surpasses Sora for realism and integration into creative workflows.
Still, VEO has not fully reached what agencies would consider production-ready quality. Marketers say that, like other leading tools, it can still exhibit AI sheen and falls short of the fully polished, final output needed for top-tier campaigns.
Sora: B
OpenAI’s Sora, a text-to-video model, reshaped perceptions of generative video when it arrived in December 2024. Its major update, Sora 2, launched in September on an invite-only basis for iOS users and introduced realistic sound effects. Earlier this month, Disney and OpenAI announced a partnership that allows Sora users to incorporate characters from Star Wars, Pixar and Marvel franchises in their content.
Marketers credit Sora with strong consistency in characters, environments and other visual elements. However, they repeatedly point to its AI sheen as a weakness.
“Sora had a really nice cinematic quality to it. I find the lighting and the idea of cinematic look to be great there, but it has a real uncanny valley with skin textures and things like that,” said an agency group creative director, who spoke on background.
Despite the high-profile Disney deal, some marketers say Sora has been overtaken by VEO in terms of hyperrealism and ease of integration into broader creative pipelines. Concerns also remain about Sora’s training data, copyright exposure and intellectual property usage.
Midjourney: B
Midjourney, launched in 2022, is one of the earliest widely adopted tools for AI-generated visuals. It has released frequent updates to improve realism and user control. At the same time, it faces lawsuits from companies including Disney and Universal, prompting some marketers to question its suitability for commercial work.
While often associated with highly stylized, “Marvel movie-style images,” according to the first creative director who spoke on background, Midjourney is viewed by many agencies as unreliable and commercially risky due to questions around its training data.
“Midjourney could benefit from more control,” said the agency group creative director. “Midjourney has been a little bit more of that Wild West slot machine — put in a prompt and here’s four very different results.”
As a result, its position in agency toolkits has weakened relative to Nano Banana and Adobe Firefly. At least one agency now labels Midjourney a “no-no” for client-facing work.
No single dominant tool
Marketers emphasize that there is no single platform that meets all their needs. Many teams combine several tools across image and video workflows to balance strengths and weaknesses in realism, control, integration and legal risk.
Overall, they are still waiting for generative tools that deliver reliable consistency, granular control and deep integration with existing systems. While the promise of AI is to speed up, personalize and scale content creation, the current landscape remains fragmented.
“Consistency is probably the number one thing that you’ll need to give people,” said Jon Weidman, svp of development at Wavelength, an entertainment production company. “Until that challenge is addressed, all of the platforms should be running at it to try to solve it.”