ChatGPT + Stable Diffusion: Part 3

It’s been a long day, and I’m tired. I don’t want to have to work very hard right now.

But I do want to to see a beautiful and imaginative landscape, inspired by Studio Ghibli.

So what do I do? First, I ask ChatGPT, one AI app, to generate a Studio Ghibli-inspired scene. Then, I feed the scene into CHARL-E, another AI app that generates an image based on those words.

Here’s the bare-bones prompt I use: “I want you to create a new plot in the same style as Studio Ghibli.”

I copy-and-paste the first paragraph generated by ChatGPT:

The scene takes place in a lush forest, with tall trees reaching up to the sky. Sunlight streams through the canopy, dappling the forest floor with patches of light. In the center of the scene, there’s a small stream that runs through the forest, bubbling over rocks and winding its way through the trees. Along the banks of the stream, there are small patches of wildflowers and ferns. In the distance, there’s a large hill that rises up from the forest floor, with a small village nestled at its base. Smoke rises from the chimneys of the village houses, indicating that the inhabitants are going about their daily lives.

…and feed it into CHARL-E to generate this image:

Isn’t that beautiful? No small village in the distance, but I can practically hear the babbling of that creek.

Now, I repeat the process, but this time I turn to Disney for inspiration. My prompt is “I want you to describe a single image using Disney as artistic inspiration.”

I do have to edit down the generated scene to remove dialogue and action, even though I asked for a description of a single image. I find that I also have to append “In the cartoon style of Disney” to achieve the desired aesthetic from CHARL-E.

A beautiful garden bursting with vibrant colors and lush foliage. A gentle stream trickles through the center, surrounded by a small stone bridge and delicate wildflowers. In the cartoon style of Disney.

Image, courtesy of CHARL-E:

The style is cartoonish, but not clearly Disney-esque. However, the stream, stone bridge, and wildflowers are present, none of which I had to dream up myself. Those details came from ChatGPT.

These are only single scenes and still images, but with a little extrapolation, a tantalizing possibility lies before us. One can envision a future where entire movies, with intricate plots and vivid imagery, can be generated from a few simple keywords specifying the genre and style for the evening’s entertainment.