zoom_in Click to enlarge

Parti

▲ ▼

description Parti Overview

Parti (Pathways Autoregressive Text-to-Image) is a text-to-image artificial intelligence model developed by Google Research and introduced in 2022. Unlike diffusion-based models that generate images iteratively from noise, Parti treats image generation as a sequence-to-sequence translation task using an autoregressive transformer. It functions by converting input text into a sequence of visual tokens based on its training data. The model provided AI researchers with a distinct architectural approach to image synthesis, demonstrating strong capabilities in visual reasoning and complex prompt adherence.

help Parti FAQ

How does Parti's image generation approach differ from diffusion models like Stable Diffusion?

Parti treats text-to-image generation as a sequence-to-sequence task using an autoregressive transformer, predicting image tokens one at a time from the text input. This contrasts with diffusion models, which iteratively denoise a random signal into a coherent image over multiple steps.

Who developed Parti and when was it introduced?

Parti was developed by Google Research and introduced in 2022. The model was described in a research paper focused on scaling autoregressive architectures for content-rich text-to-image generation.

What image tokenization method does Parti use?

Parti uses a ViT-VQGAN image tokenizer to convert images into discrete visual tokens that the autoregressive transformer can process. This allows the same seq2seq framework used in language translation to be applied to the image-generation problem.

What model sizes did Parti experiment with?

Parti was tested at multiple scales, with the largest reported version using around 20 billion parameters. The research showed that increasing model size improved the system's ability to follow complex, content-rich prompts with multiple objects and attributes.

swap_horiz

Looking for Parti alternatives? Compare top competitors ranked & scored

arrow_forward