Generating Images for Dreams with Artificial Intelligence using the Temenos Dream App

Introduction

At Temenos Dream we see a lot of potential for artificial intelligence to improve the way we interact with our dreams. Dream analysis takes a lot of work and creativity. But AI (artificial intelligence) can help us bring our dreams alive faster. AI is just another tool we can use like any other dream analysis technique to get closer to the meaning of our dreams. One way Temenos Dream uses AI is for generating images of our dreams. This is not only super cool! But it helps us convey our dreams to others so they can get a more accurate understanding of what we dreamt.

Quick Start (only 3 clicks)

It is 3 button clicks to generate your first image with artificial intelligence. Check out the image below and follow these steps:

Go to the dream page where you wrote your dream
Scroll down and click Add Image
Click Generate Prompt which will have an AI create a summary of your dream description which will be used as a prompt to send to the image AI.
Then click Generate Image
Once you get your first image tap on the prompt to edit it and click Generate Image again to get another image. That is the basic process!
Delete any images you don't like by clicking the trash icon on the image
Download images for sharing on social media by clicking the arrow on the image
Read on to learn more

What is under the hood

The way AI image generation works is like a statistical funnel. Text goes into the top of the funnel and pixels come out the bottom. The AI has been shown millions of categorized images from the internet. Then it is told to recreate an image from a text based description. During the training process another AI checks if the image is too similar to any of the original images. If it is too similar then it tries again. It keeps doing this over and over until it can reliably create completely original images that are not in the original training set. The AI isn’t really thinking like a human would. It is just taking a text description and statistically mapping that to a 262,144 pixel array aka an image.

How we can generate better images

Now that we know that the AI understands images on the internet and works via statistics, how can we use that to our advantage? Here are a few things that might help.

Use words that humans and the internet have a lot of information about

For example I had a dream about a mule (a cross between a donkey and a horse) but horses are much more common and they look similar to mules. In fact in the dream I wasn’t even sure if it was a horse or a mule. If you search on the https://trends.google.com/ website you can see that there are about 9 times as many searches for “horse” than there is for “mule”. You can also check this website to see which images the AI has been trained on: https://haveibeentrained.com/ … So in this case it gave the AI much more data to pull from when I used the word “horse" instead of the word “mule” in my image prompt. Another example is “Temenos”. Temenos is not a commonly used word so it would be better to use something like “temple” or “sacred space”.

Describe yourself and other people in your dream

The AI doesn’t know who you are! Unless you are famous. So when you want to include yourself in the dream you have to describe yourself. For me I would say: “a young man”, or “a young white man with shaggy brown hair” but you have to be careful with including too much information because it might pick up on certain words too much. For example if you say “a young man with shaggy brown hair” and you also have the word “carpet” or something in your prompt it might make the carpet shaggy and brown or it might even make the carpet look like hair. Which leads us to the next point.

Withhold information that is not crucial or has a complex definition

You have to make sacrifices. Not everything can be represented accurately together. Think about your dream as a word cloud (shown below) with the most important symbols being the largest. Then think about which words the AI would know most about. Use those words and any necessary context but nothing else.

Use a style

There are many styles of images the AI is trained on. You can tap into certain genres of images by adding a style or even a famous artist name to the end of your prompt such as “old photo”, “anime”, “computer render”, “van gogh” etc.. A couple of my favorites are “unreal engine 8k”, and “hd photo high quality”. One of my dreams reminded me of an anime so I used that style and it worked perfectly to capture the dream.

Avoid triggering the sensitive image filter

The AI has a filter to prevent it from returning pornographic images. If you try to generate an image and nothing happens it might be because you triggered the filter. One thing that seems to trigger it is if you mention someone is young. Words like “young girl”, “young boy”, “child”, and “kid” get blocked sometimes. If you get blocked just change some words and you should be good to go.

Learn by practicing

Even though we have AI to do the heavy lifting, generating images is a skill that takes time to perfect. Keep testing different methods and find what works for you! Thanks for reading!

For more specific help contact us at support@temenosdream.com