What does a professionally designed ad actually look like, across brands and industries? This dataset holds 35 finished ad creatives, one for each of 35 unique, synthetic brands across 4 industries, e-commerce/retail, food and beverage, beauty/wellness, and tech, all delivered as social ads. We gave designers from the Contra network a simulated client brief for each brand, including:
- Company name
- Brand guidelines
- Logo
- Product image
From there, we asked creative professionals to create a final, on-brand square ad that a brand would actually post to a social feed. Designs were created in Figma, incorporating all brand guideline features and inputs.
The published dataset is the brand input and final output. Every brand is represented once, so the set spans 35 distinct brand identities and use cases, distributed across e-commerce/retail, food and beverage, beauty/wellness, and tech brands
The range of visual conventions, including product-forward retail layouts, appetite-driven food imagery, clean beauty aesthetics, and information-dense tech messaging represents the range of ad creative being commissioned by real companies to be created by real designers. This structure makes the data directly usable in several training contexts:
- Text-to-image fine-tuning: a curated corpus of real, on-brief square ad creatives for conditioning image models on professional layout, type, and brand style
- Layout & text-in-image generation: real examples of type hierarchy, logo placement, and copy integration
- Benchmarking & evals: a human-designed reference set for measuring generated ads against professional output, by industry
The dataset is released under the CC BY 4.0 license on Hugging Face.