When you make AI art, you look into the best ways to craft an image prompt that gives you a result that is close to your vision. In the event that you were ever curious… there are a lot of images you can’t use. It’s quite a difficult task, despite the fact that you can get amazing results just from typing a few words into a computer.
I recently asked some AI chat software what the best practices for creating a generative AI image are, and was told by ChatGPT some weird stuff that didn’t make sense to me. (I’m guessing I worded my questions in such a way that the chatbot thought I was intending to create an image AI of my own instead of using one for image creation)
But asking Google’s Bard gave me results that I understood much more:
- Be specific. The more specific you are in your prompt, the more likely Midjourney is to generate an image that matches your vision. For example, instead of saying “a beautiful landscape,” say “a mountain range with snow-capped peaks and a river running through it.”
- Use keywords. Midjourney can understand keywords, so use them to your advantage. For example, if you want to generate an image of a cat, you could use the keywords “cat,” “feline,” and “domestic.”
- Use adjectives and adverbs. Adjectives and adverbs can help to add detail and realism to your images. For example, you could say “a realistic painting of a cat sitting on a windowsill” or “a stylized illustration of a cat wearing a hat and sunglasses.”
- Use negation. You can use negation to exclude certain elements from your images. For example, you could say “a cat without a collar” or “a landscape without any people.”
- Use filters. Midjourney offers a variety of filters that you can use to change the style of your images. For example, you could use the “oil painting” filter to make your images look like they were painted by hand.
- Use creativity. Don’t be afraid to experiment with your prompts. The more creative you are, the more interesting your images will be.
However, this is not always what I have found works for me in practice.
For me being specific doesn’t usually work… at a certain point, adding more information to Midjourney or DALL*E or whatever AI you use doesn’t seem to change anything all that much.
This is a simple photo of a baboon with a balloon that I created with the Midjourney AI. It’s not a very impressive balloon either but I tried to add to the idea.
I tried to create an image of the baboon holding a balloon and a torch!
First off, the AI seems to have merged the items. Most of the images appear to show either a balloon with a flame or a balloon with a candle inside or just a candle. A candle is not a torch but even if it was, the image would still be off because a balloon torch isn’t a thing!
Well, that didn’t work… at all. I mean, they are still very good… better than what I could do… they look fairly realistic, if not “Photoshopped”… but the light actually does seem to exist in both scenes so that is cool.
But being specific didn’t help here and I’ve found that it doesn’t always help when creating AI images. And that is usually the case with any of the practices listed above… you can’t really rely on the programs to get your vision 100% right… isn’t that what makes it art though?
Personally, I think the fact that not every element is perfect is what adds some flavor to the image… which may not be what others want for their own art, articles, stories or whatever someone intends to use an AI image for. So I think AI image generators will ultimately be able to create exactly the picture in your mind, but for today… anyone using AI art has to have a lot of patience and willingness o keep trying… failing that, you just need the ability to look past perfect and realize that it’s not just about that!
Something that does work though, is using modifiers… like writing realistic or comic book style… writing in a setting usually works, but sometimes is just not gonna produce the exact results you want!
See his balloon got better when I put him in the subway!