AI advertising platform Omneky says its creative-generation workflow achieved the highest score in TasteBench, an evaluation comparing advertising outputs from its system with six configurations using ChatGPT and Google Gemini. Omneky scored 8.8, compared with 7.6 for the next-highest configuration, ChatGPT using GPT-6 Astra at medium reasoning.
The evaluation covered 52 advertising requests across four brands, generating 364 ads for assessment. According to Omneky, ChatGPT configurations scored between 7.3 and 7.6, while Gemini configurations scored between 6.8 and 7.2. The results suggest that how an AI model is configured and directed can influence the quality of advertising creative it produces.
The distinction matters for marketers adopting generative AI. Rather than evaluating a model in isolation, TasteBench compares complete workflows, including the software that supplies brand context, guides creative decisions and selects image-generation tools. Omneky’s results therefore measure its advertising system as a whole, not simply the underlying model’s capabilities.
How Omneky’s Advertising Harness Works
An AI harness is the software layer that structures a model’s task, supplies relevant information and guides its output. In Omneky’s advertising workflow, this includes interpreting brand information, developing art direction, incorporating product and logo references, and choosing an image model.
These steps address a practical challenge in AI-assisted advertising: generating an image is not the same as producing a usable brand advertisement. Campaign assets must align with visual identity, product presentation and creative direction. A workflow that incorporates those requirements may reduce the amount of manual correction needed before an asset is ready for review.
For agencies and in-house marketing teams, this approach points to a broader shift in advertising AI. Competitive differentiation may increasingly depend on orchestration, brand-specific context and quality controls alongside the capabilities of the underlying generative models.
Omneky also reported a separate readiness result. Sixty percent of its ads passed every applicable benchmark check, compared with 13% for the next-best configuration. This metric measures compliance with the evaluation’s required checks, rather than campaign performance or commercial effectiveness.
Market Landscape: Evaluating AI Beyond Image Generation
The advertising industry is testing generative AI for creative production, campaign variations and faster asset development. Yet model benchmarks do not necessarily reflect how well a system performs in a commercial advertising workflow.
TasteBench’s approach focuses on the finished creative generated by each configuration. Omneky said three anonymous judges assessed the outputs using the same standard. The ChatGPT and Gemini tests used APIs with web search and image-generation capabilities, while checks using the consumer-facing applications reportedly produced similar results.
The evaluation offers a comparison of creative quality under the tested conditions, but several limitations matter. The sample covered four brands and 52 requests, and the announcement does not establish whether the results generalize to other industries, campaign formats or larger creative workloads. It also does not provide enough detail to independently assess the scoring rubric, judge agreement or statistical significance.
The findings therefore should not be interpreted as proof that Omneky consistently outperforms every general-purpose AI system in production.
Strategic Outlook
For advertisers, the central question is shifting from which AI model can generate an image to which complete system can produce brand-appropriate creative with less rework. A specialized harness can combine brand assets, instructions and model selection into a repeatable process, potentially making creative generation more consistent across campaigns.
For agencies, these workflows could help scale creative development while preserving human oversight. Publishers and media buyers may benefit indirectly if advertisers can produce more campaign variations for different placements, although the benchmark does not test delivery performance or media effectiveness.
Omneky’s results also highlight the need to evaluate AI advertising tools against practical requirements, including brand consistency, factual accuracy, logo and product fidelity, review time and cost per approved asset. The company has demonstrated a strong result in its reported benchmark, but campaign-level testing will be needed to determine whether that advantage translates into business value.
Top Insights
- Benchmark leader: Omneky reported an 8.8 TasteBench score, 1.2 points above the next-best tested configuration.
- Workflow matters: The evaluation measures the complete advertising harness, including brand context and creative direction.
- Higher readiness score: 60% of Omneky’s ads passed all applicable checks, versus 13% for the next-best configuration.
- Limited sample: The test covered four brands and 52 requests, so broader applicability remains unproven.
- No ROI evidence: The evaluation measures generated creative, not advertising returns, conversions or campaign performance.
Get in touch with our Adtech experts
