Would a curated dataset of ~4000 social media design layouts be useful for training or fine-tuning design models?

🤗Datasets
1

I’m a graphic designer who has created around 4000 social media posts over the past couple of years.

Most of them follow common social media layout structures used for community engagement and announcements, such as:

  • headline + visual + CTA

  • centered quote layouts

  • announcement cards

  • question or engagement posts

  • festival greeting posts

The designs follow social media composition patterns (text hierarchy, visual balance, spacing, etc.).

I’m thinking about organizing them into a structured dataset with metadata such as:

• layout type
• post category (engagement, announcement, greeting, etc.)
• text content
• basic layout structure

My question is:

Would a curated dataset like this (~4000 samples) be useful for training or fine-tuning models that generate social media layouts or designs, or would it generally be considered too small to be useful?

I’m curious about the usefulness of domain-specific layout datasets compared to much larger but more general image datasets.

Any insights would be appreciated.

1 Like
2

fine-tuning models that generate social media layouts or designs

A dataset of 4,000 well-labeled images (or some layout information) should be sufficient for fine-tuning.

However, finding a model that outputs SNS-like layouts to use as the base for fine-tuning might be a bit of a struggle. There don’t seem to be many existing ones.