How to Generate Ghibli-style Images with AI

OpenAI just released Image Generation for their latest GPT-4o model and the results are impressive. It unlocks a new level of precision for taking input images / prompts and generating output images that both align to the inputs and look good.

While you might imagine that people would use super intelligence to solve the world's most pressing problems - in reality most people just want to create funny images.

In this post we'll walkthrough using GPT-4o Image Generation to generate Studio Ghibli-like illustrations of your own images.

Generating Studio Ghibli-styled Images

GPT-4o Image Generation is currently available to Plus ($20 / month), Pro, and Team users of ChatGPT. There are plans to roll it out to Free tiers as well but Sam Altman (OpenAI CEO) said it may be a few weeks as the feature was unexpectedly popular.

To generate an image in Ghibli-style:

• Login to ChatGPT and ensure you have the appropriate plan
• Select ChatGPT 4o in the top left drop down
• Find an image you want to use and upload it to the chat
• Type something like Draw this image in the style of Studio Ghibli

Wait until the image generation is done (~30 seconds) and either download it or ask it to try again (maybe with a different prompt) if you don't like it.

Limitations

• Generation takes awhile - 30+ seconds. Will probably get faster as demand dies down and they improve infrastructure.
• Copyrighted Materials - Can't draw copyrighted material like Pokemon although the controls seem spotty cause some brands go through. In general it can draw an image you have in a different style if you don't specifically tell it to generate something with copyrighted contents.
• Some hallucinations / bad outputs - May need to regenerate an image several times with different prompts.

Also note: API access for Image Generation is not yet available - with estimates being a few weeks before that happens.

Next

I've been having fun generating images with AI on and off for the past few years but I have to admit that this new release really improves on the precision and quality of image outputs.

An exercise I've enjoyed with this model is taking some of my favorite IG photos, running them through the AI, and then sending them to family and friends.

If you liked this post you might also like:

• How to call Open AI's API with F# (GPT-4o)
• Building an in-browser Auto-clicker with JavaScript
• Were Tesla’s Optimus Robots actually Autonomous (or Remote-Controlled by Humans)?