Turn any idea into an image in seconds with Midjourney AI!

Name: Turn any idea into an image in seconds with Midjourney AI!
Uploaded: 2022-12-24T05:57:56.000Z
Duration: 38 min 58 s

Introduction

The video introduces the topic of generative AI and how it has changed the way we perceive art. It explains that the video will show how to use Mid-Journey, a publicly available generative AI algorithm, to create avatars and images.

Generative Adversarial Networks (GANs)

GAN is a technique developed by Ian Goodfellow in 2014.

GAN pits two algorithms against each other - a generator and a discriminator.

The discriminator is trained on billions of images to extract a mathematical representation called an embedding.

Diffusion algorithms take this idea further by training on how noise from billions of images looks like and then applying specific algorithms to clean up noise from images.

Steps for Generating Images with Mid-Journey

Step 1 - Take billions of images on the internet with text descriptions of what's in them.

Step 2 - Slowly add noise to the images around a thousand times.

Step 3 - Show noisy images to an AI and teach it step-by-step how to remove noise from them.

Step 4 - Generate noise that matches the mathematical representation of an image.

Step 5 - Clean up the noise, getting a small but clean image.

Step 6 - Upscale the image.

Using Mid-Journey to Create Avatars

Mid-Journey version 4 is more powerful and capable than version 3.

The video provides a link to the Mid-Journey channel on Discord for examples of how to use it.

Introduction to MJ Version 4

In this section, the speaker introduces MJ Version 4 and explains how it generates images.

How MJ Version 4 Works

By default, MJ Version 4 generates four images and creates corresponding buttons on the bottom of the screen.

The "U" button upscales an image while the "V" button creates variations on that image.

To prevent something from appearing in an image, add "--no" before the object you want to remove.

Experimenting with Styles

In this section, the speaker discusses how to experiment with different styles when generating images using MJ Version 4.

Using Tags for More Predictable Results

Adding tags that people are likely to use when searching for photos can result in more predictable results.

Examples of Different Styles

Using tags such as "realistic," "8K," and "no clouds" can create a realistic photo of a boat sailing into the sunset.

Adding a style such as "dull" or "Leonardo da Vinci" can incorporate elements of classic art into an image.

Pop culture references such as Superman or Homer Simpson can also be incorporated into an image using different styles.

Controlling Image Quality and Weighting Words

In this section, the speaker explains how to control image quality and weighting words when generating images using MJ Version 4.

Controlling Image Quality

Use "--Q2" to make the AI work twice as hard on an image for more detailed results. However, overworking an image can make it look less realistic.

Weighting Words

Each word used in generating an image can be given a weight by adding "::number" after the word. For example, "Flying Pig::0.5 Cloud::2" will result in an image with more clouds than flying pigs.

Generating Images with Mid-Journey

In this section, the speaker explains how to generate images using Mid-Journey. They discuss how the generative algorithm works and how to change the aspect ratio of an image. They also explain how to tweak specific parts of an image by using a seed.

Changing Aspect Ratio

The generative algorithm generates blocks of pixels that create an image.

You can change the aspect ratio of the image with iPhone-AR-free: 2 will generate a wide image, 2:3 will generate a tall image.

Generate both types of images to see the difference.

Tweak Specific Parts of an Image

When generating an image, many parameters are generated randomly.

There is no such thing as random in computers; all kinds of algorithms come up with pseudo-random numbers that influence various aspects of the generated image.

If you want to tweak just a small part of an image, use a seed.

Next to the message containing your generated image, there is a little smiley face icon. Click on it and within seconds, Mid-Journey will send you a message with all information about that specific image, including its seed.

Copy and paste this seed into another prompt for Mid-Journey to generate another similar but not identical version of your original prompt.

Using Other Images as Inspiration

You can use other images as sources or inspiration for your generated images by giving Mid-Journey their URL and adding text prompts.

To upload an image easily, drag and drop it into the chatbox and submit it in a separate message so you can copy its link later.

Every detail in your uploaded picture will have some impact on your final result. For example, if there's a bit of white on the side, you'll see that white in the final image.

If Mid-Journey generates an image that's close to what you want but doesn't look like you, feed it its own output as the source for the next image together with your own picture. Don't forget to add descriptors to your prompt.

Generative AI Algorithms: Hands and Faces

In this section, the speaker discusses how generative facial algorithms can be used to fix distorted faces and restore old photos.

Generative Facial Pile (GFP)

GFP is an algorithm that understands how a face looks like and fixes any issues with the face.

It was designed to restore old photos but works well on AI images too.

To use it, upload the image to Jeff with the URL provided in the description. All you need is a GitHub account, and within seconds, you get a fixed face that looks much better than the original.

Using Me Jonah for Design Ideas

In this section, the speaker explains how Me Jonah can be used to generate icons, dashboards, landing pages, and design ideas.

Generating Icons

Me Jonah can be used to generate icons for your app.

The generated icon is beautiful even after upscaling.

Beautiful Dashboards

Me Jonah can also be used to generate beautiful dashboards or user interface design ideas with a simple prompt.

The dashboard generated by Me Jonah is stunning.

Landing Pages

You can ask Me Jonah to generate a beautiful landing page using hyphen hyphen ar322 in order for it to be landscape instead of square.

The landing page generated by Me Jonah is gorgeous.

Bonus Tip: Recreating an Image Pump from an Existing Image

In this section, the speaker provides a bonus tip on how to recreate an image pump from an existing image.

If you have an image that is almost what you want but don't know the pump of it, you can use pound to IMG.

Upload the image and submit it, and within seconds, you get a prompt that can generate specifically that image or something very close to it.

The generated prompt is a foggy park with benches and tilt-shift photo by Mike Pilk Shutterstock contest winner tonalism soft mist high dynamic range and mist.