Google's Veo video generation tool significantly outperformed OpenAI's Sora in a head-to-head test by a major advertising campaign, delivering superior 4K quality and scene extension capabilities. However, the real story reveals a sobering truth: even the best AI video tools today require brutal human oversight, with creative teams discarding over 400 video outputs to produce just 10 campaign-ready ads.

Wonga, a South African fintech company offering short-term loans, set out to extend its 2026 "Yoh! to Yebo" marketing campaign across television, video-on-demand, digital platforms, and out-of-home advertising. Rather than rely solely on traditional production, the team decided to test AI video generation as a way to create nimble, cost-effective short-form content that could complement their existing TV spots. The experiment compared three major AI platforms: Sora from OpenAI, Grok Imagine from X, and Veo from Google Gemini.

Why Did Google Veo Win Out Over Sora?

Veo emerged as the clear winner for several practical reasons. The model delivered consistently high-quality 4K video output and offered robust scene extension capabilities, allowing the creative team to gradually expand eight-second clips up to a full minute. Google's proprietary Flow platform also enabled the team to extend or remaster scenes without immediately heading to the editing booth, streamlining the workflow in ways the competing tools could not match.

Yet the victory came with a massive caveat: the team discarded more than 400 video outputs before landing on 10 that were campaign-ready. This staggering rejection rate underscores a fundamental limitation of current AI video generation technology. As Bryan Smith, Wonga's digital media manager, explained, the challenge goes far deeper than raw output quality.

What Are the Real Barriers Holding AI Video Back?

AI video generation tools struggle with several critical creative and technical hurdles that no amount of computing power has yet solved. Understanding these limitations reveals why human creativity remains irreplaceable in advertising and content production.

• Consistency Across Scenes: Characters, locations, and visual elements rarely remained consistent from one generation to the next. Faces would change between shots, outfits would swap unexpectedly, and new props would appear without being prompted, forcing the team to abandon multi-scene narratives and focus on single-scene outputs instead.
• Content Filtering and Edginess: AI models can only draw from what they have been trained on, and built-in modesty restrictions steer them away from the absurd, edgy territory where the best advertising lives. The more original and funny the concept, the harder the tool fights back, making it nearly impossible to generate truly novel creative ideas.
• Dialogue and Audio Quality: The team eliminated dialogue entirely from their outputs because AI-generated speech remained unreliable. Sound design and foley audio, which can transform a viewer's understanding of a scene from a chuckle into a laugh, still require human expertise and multiple platforms to execute properly.
• Local Cultural Nuance: AI cannot judge what is "good enough" for a specific market or audience. Bringing humor to life with care and taste, ensuring that concepts that poke fun do not lose their meaning or offend, requires human judgment and deep cultural understanding.

One particularly telling example illustrates the absurdity of the current state: when the team tried to prompt an AI to generate a scene about a flooded apartment, the model abandoned physics entirely and rendered household objects floating through the air like a scene from Star Wars, rather than floating on water.

How to Use AI Video Tools Effectively in Creative Work

• Plan for Massive Rejection Rates: Expect to discard the vast majority of outputs. Generation is cheap enough to fail fast, but only if your process anticipates a brutal hit rate rather than a tidy one. The Wonga team kept 10 clips from more than 400 attempts.
• Think in Single Scenes, Not Stories: AI loses narrative continuity across cuts, so build short, single-scene beats and stitch them together manually rather than asking the model to carry an entire story arc. This dramatically improves consistency and creative control.
• Invest in Quality Prompting and Direction: Having a clear vision is essential, but recklessly worded input can produce bizarre results. A director with strong prompting skills can guide the AI toward usable outputs, though even expert prompting yields mostly unusable results.
• Keep Talented Teams, Not Replace Them: AI compresses the distance from idea to frame like never before, but it cannot judge quality, land local nuance, or make an audience feel emotion. Treat AI as an accelerator for your team, never a replacement for it.

The Wonga team's conclusion was unambiguous: AI is not ready to develop full-blown television advertisements that replace human hands. Brands looking to invest in creative AI capabilities should match that investment with capable teams or agencies to enhance their executions, not replace them.

This real-world test offers a crucial reality check for the AI video generation industry. While tools like Veo represent genuine progress over their predecessors, the gap between "impressive" and "production-ready" remains vast. The future of AI in advertising is not about replacing creative professionals; it is about giving them faster, more scalable tools to amplify their ideas. But only when those tools are paired with human judgment, cultural expertise, and editorial oversight.