Hugging Face releases Palette, a small AI model for custom slide deck design

Hugging Face has introduced Palette, a specialized 20B model designed to create slide decks that follow a specific design language1. The tool addresses the tendency of larger, frontier AI models to default to generic styles and ensures that content remains within a user's own infrastructure1.

The model functions by separating the design process into two distinct roles: a designer that plans the deck's narrative and a coder that executes the precise placement of elements1. This approach is intended to prevent common errors in small models, such as overlapping shapes or text extending beyond slide boundaries1.

What we know, and what we don't

Hugging Face claims that Palette outperforms other small models by using synthetic data trained on a specific design language1. The company reports that the model ranks first in two of the three domains measured on PresentBench, a benchmark for presentation generation1.

The model relies on a harness to check its output, as small models often struggle with the arithmetic required to fit text and shapes accurately1. It remains unclear how the model will perform across diverse design environments or if it will be accessible for broader use1.

Why it matters

For users, Palette offers a way to automate slide creation while keeping sensitive data private and ensuring the output matches a company's specific visual identity1. By running on local infrastructure, it removes the need to rely on external systems that may not support custom templates1.