The Math Behind Ghibli-Fication.

Author(s): Surya Maddula

Originally published on Towards AI.

OpenAI released a new feature for text to image generation with GPT 4.5, and the entire internet has been going absolutely crazy with creating their images and some iconic memes into the style of Ghibli-Anime, and the results are shockingly amazing!

An Iconic Meme in Ghibli Style

The Ghibli generator uses diffusion models, which are a type of Generative AI that learns to reverse a noise-adding process to create images from text prompts.

GPT-4o was trained on a vast dataset which might’ve included Studio Ghibli images which were available online. This training is basically what enables the model to recognize and reproduce the Ghibli style when prompted with phrases like “in the style of Studio Ghibli” because otherwise the model doesn’t understand what “Ghibli style” means.

You can try this yourself if you have a ChatGPT paid plan by uploading your image and typing in “Show me in Studio Ghibli style.” The model then generates images that capture the soft, painterly textures and warm, natural color palettes characteristic of Ghibli art.

Diffusion models are literally the heart of the Ghibli generator’s functionality. These models work by adding noise to images iteratively and learning to reverse this process, which then denoises to generate new images from… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI