contralabs/creative-ad-design-dataset

Ad Creative Design Dataset (Preview)

This is a preview release of finished ad creatives produced by professional designers working from complete brand briefs. Each of the 35 rows is one fictional consumer brand: the designer received the brand's written guidelines, its logo, and a product photograph, and delivered a square social ad creative that was reviewed and approved. The row carries all three images alongside the brief broken out into structured fields: personality, color palette with hex values, typography rules, ad archetype, and audience. Most open ad-creative corpora are scraped finals with no record of the constraints the designer worked under. This one keeps the full input → output pair: the complete brand system as the model-facing brief, and the approved human deliverable against it.

From Contra Labs

This dataset is a sample of the human data Contra Labs builds on demand. We are an independent human-data and creative-evaluation lab, backed by a network of verified creative and domain experts. We produce expert evaluations, rankings, benchmarks, and custom datasets like this one for AI teams.

Other datasets from us on Hugging Face:

• —Descript video-editing computer-use trajectories. Long-horizon editing sessions with narration-grounded reasoning.
• —Adobe Premiere editing trajectories. Native desktop-app editing sibling to the Descript set.
• —Video detail annotation. Dense human descriptions and annotations of video content.

Browse all our datasets on Hugging Face.

Want data like this for your domain? We build expert-made creative deliverables, pairwise preference data, and expert evaluation scores to your schema and difficulty. Reach out at contralabs.com or partnerships@contralabs.com. See Working with Contra Labs for more.

What's in it

35 brands, one row each, in a single Parquet file with all images embedded:

Notable properties:

• —Complete input → output pairs. Every ad creative comes with the exact brief the designer worked from: the written guidelines plus the logo and product image they were required to use. Nothing about the brief has to be reconstructed after the fact.
• —Fully specified brand systems. Each brand's palette names the same six roles (Primary, Secondary, Background, Surface, Text, Accent), each with a color name and hex value, alongside typography rules (family, casing, weights) and a brand personality.
• —Approved professional work. The creatives are deliverables by professional designers that passed review; the dataset contains only the approved final per brand.
• —A small archetype taxonomy. Each brief assigns one of five ad archetypes, and ten briefs add an Urgency or Seasonal modifier, so outputs can be compared within a controlled brief structure.
• —Fictional brands throughout. All 35 brands, logos, and products were created for this work and do not correspond to real companies.

The task (study setup)

Each designer received one brand's brief: the brand-guidelines.txt (personality, product, six-role color palette with hex values, typography rules, ad archetype, category/audience, and occasionally packaging direction or logo-usage notes), the brand's logo, and a product photograph. The task was to design a square social ad creative for that product, honoring the palette, typography, archetype, and logo-integrity instructions in the brief. Finished creatives went through review, and this dataset contains the approved deliverables.

Quickstart

from datasets import load_dataset

ds = load_dataset("contra-labs/ad-creative-design-dataset", split="train")
row = ds[0]
print(row["brand"], "-", row["product"])
print(row["color_palette"])
row["ad_creative"].show()

Each row is self-contained. The three image columns are Hugging Face Image features with bytes embedded, so no file resolution is needed.

The structured brief columns were parsed from each brand's original brand-guidelines.txt, which the parquet reproduces in full across its fields. The parquet is the sole distribution: images are embedded, and there are no loose files to resolve.

Intended use

• —Training and evaluating image models that generate ad creatives conditioned on brand assets (logo, product shot) and written brand guidelines.
• —Judging generated creatives against the same briefs the human deliverables satisfied; the approved creative serves as a professional reference point per brief.

This is a preview rather than a training-scale corpus: 35 briefs across one deliverable format (square social ad creative). It is intended for evaluation and qualitative study rather than large-scale training.

Working with Contra Labs

Contra Labs is an independent human-data and creative-evaluation lab, backed by a network of verified creative and domain experts. This dataset is one example of our work. We partner with AI teams on:

• —Evaluation, rankings, and benchmarks. Expert human judgment on model outputs across text, image, video, audio, UI, and multi-modal work, scored for quality, style, and brand fit.
• —Custom dataset creation. Expert-made creative deliverables, computer-use trajectories, pairwise preference data, and evaluation scores, built to your domain, schema, and difficulty.

To commission an evaluation or dataset for your domain, reach out through contralabs.com or email partnerships@contralabs.com.

Provenance

The creatives were produced by professional designers from Contra's network as commissioned work. All brands, logos, products, and briefs are fictional and were created for this dataset; they do not depict real companies or real products.

Citation

@misc{contra_labs_ad_creative_design_2026,
  title        = {Ad Creative Design Dataset (Preview)},
  author       = {Contra Labs},
  year         = {2026},
  howpublished = {Hugging Face Datasets},
  note         = {Preview release}
}

License

Released under CC-BY-4.0. Free to use with attribution to Contra Labs.