How to create an image with AI when you are not an artist?Let’s say that you are updating the website Everygame mobile casino and you want to add some original images, but you are not an artist. How exactly does that work?
Well, I decided to try to figure that out.
What website creates images using AI?
Here are some examples of websites, but this is not a complete list:
- DALL-E-2 — Created by OpenAI
- Midjourney — Midjourney is a generative artificial intelligence program and service created and hosted by a San Francisco-based independent research lab Midjourney, Inc. Midjourney generates images from natural language descriptions, called “prompts”, similar to OpenAI’s DALL-E and Stable Diffusion.
- DreamStudio — It is prompt engineering. It’s just using words to paint a picture of what you want to create.
- Getimg.ai/text-to-image — It also uses a prompt, but you also have to use a control image.
Example using DALL E 2
Generate images from a description in natural language.
Well, this test did not get very far. A person cannot even test DALL E 2 without buying credits, and I have no desire to buy credits if I have no idea if it is going to even come close to producing images that I want.
Example using Midjourney
I created an account. I downloaded the app. I could not find anything that has anything to do with creating images from prompts. Midjourney was a waste of my time.
Example using DreamStudio
With this application, you have to start with an existing image. If I had an existing image that I wanted to start with, I would not need to use this application. Ugh.
Final test – Getimg.io
Let’s see if I have any better luck with this application.
This is another application where you have to start with an existing image. Okay, I give up. Let’s see if somehow I can make this work.
I am starting with the prompt “Create an image of people with wings as tried throughout history.” Yes, it is a dumb prompt, but it is intended to go with a simple article about the history of people trying to fly by creating wings for themselves. (It was a history lesson for an elementary school student, and I wanted an image to go with the essay).
Tip 1: Control generation with input images
“Use 7 powerful ControlNet pipelines to guide the generation using input images. Preserve depth, edges, human pose, or even turn your drawings into detailed artworks in seconds. All you need is a source image and a simple prompt.”
If I had a source image, I would not need this application in the first place.
Tip 2: Edit images with Instruct model
“Click on ‘Edit’ tab to use text-based instructions like ‘Make her smile’, or ‘Change hair color to blue’ to apply subtle edits to your images.”
This is great if you want to put sunglasses on a tiger or change the picture of your baby so it smiles. But if you are starting from nothing … not much help.
Tip 3: Try 25+ community trained models
“Be more creative by using multiple models in your art creation pipeline. Use different models to apply different styles easily without writing complex prompts.”
No idea at this point if this will help me.
Tip 4: Use AI Editor to further improve your images
“Edit photos using a few brush strokes and the inpainting model, or expand pictures beyond their original borders with outpainting.”
Anybody who thinks that AI is going to replace traditional artists is sadly mistaken. That would be like saying digit art replaced traditional paints and brushes. No it did not. It was just another set of tools for an artist to add to their toolbox. A person would still need artistic ability in order to even use this tool.
I am not an artist. I am a software engineer. I can write prompts, but I would be quite surprised if even if I could find a decent starting image I would be able to “Edit photos using a few brush strokes and the inpainting model, or expand pictures beyond their original borders with outpainting.” and make the final image look even halfway decent.
Tip 5: Create your own custom AI models
“Use DreamBooth to turn photos into a personalized Text-to-Image model that can generate stunning images of your chosen subject.”
The example shown is using an image of a boy / young man, and from that original image, creating a bunch of other images using different poses, different backgrounds, and different clothes.
In other words, you can take your favorite wedding picture and have you and your mate located on the beach, on a cruise ship, riding a roller coaster, in the forest, in the Arctic, in the city, in the country, etc. etc. etc.
So it would be more likely that “destination photography” would be taken out, rather than traditional artists. On the other hand, it still goes back to imagination. When a wedding couple is paying for a photographer, they are not just paying for a person with a camera. They are paying for their creative artistic talent. So in that sense, it will most likely increase the value of a local photographer who learns how to master these tools.
The local photographer takes a core set of pictures that the customer “loves”, and then from there, they work with the customer to create “dream pictures”.
And this website is using DreamStudio … which I was never able to get working in the last example.
Giving it a try …
Select a model … Stable Diffusion v2.1
I found an old black-and-white picture of a man with man-made wings jumping off of a cliff.
The prompt I used was “Change the image to be in color and wearing modern clothing.”
Let’s see if we can anything interesting or useful …
Talk about a great disappointment …
First of all, the AI did not convert the image from black and white into color. FAILURE 1
Second, the software did not even attempt to change the man’s clothes to modern clothes. FAILURE 2
Original Image: You cannot see the man’s face.
AI Image 1: The man’s face is clear. If that is what you are looking for, that is great.
AI Image 2: The man is wearing a bowler hat. Not what I was looking for.
AI Image 3: No hat, but a different hairstyle.
AI Image 4: Clown socks. Really? A person testing wings wears clown socks, but 1930s clothing?
Summary
Do not worry artists … AI-generated images are not even close to putting “professional artists” out of business. If you are looking for “touch ups” (getting rid of a pimple, adding a hat, getting rid of the idiot who walked in front of the camera, etc.) it can work. But if it cannot change a black and white photo to a color photo or change the clothes of the person in the image from 1930s clothing to 2023 clothing … it is not going to work.
AI-generated images are not going to go away, but in my opinion, it is going to be another tool that professional artists will use – as opposed to AI-generated images replacing professional artists.
At least that is what it looked like to me, when I tested the software in May 2023. But who knows where the technology will take us in 10 to 20 years. But in the near future, artists, you are still needed with all of your great creativity.