The search engine Microsoft Bing has a new tool called Image Creator. This tool lets you create images from text by using an AI system called DALL·E 2 from OpenAI. If you want a superchihuahua with a red cape, just type "superchihuahua with a red cape" and you get four images in just a matter of seconds (the more images you generate the longer it will take unless you pay Microsoft some money). No-one took a photo of exactly this chihuahua dressed in a red cape - it doesn't exist in real life - it was generated by the Artificial Intelligence.
Art styles
The Image Creator tool can also generate images with a specific art style, such as watercolor or anime. But there's not an overview of which art styles are available, so I thought I would try a few of them and show the result here. To get a specific art style, you type "superchihuahua with a red cape, anime art style" so it understands which art style to use. The AI is smart when it comes to generate cool images - but sometimes it's really hard to make it understand what you want based on what you think you want. For example: "superchihuahua with a red cape, anime style" generated photo realistic images, so you have to experiment a little, but the general rule is that the art style you want should come last after a comma.
2d art
Materials
If you get bored of the art style, you can generate chihuahuas made out of different materials. To get a specific material, you type "superchihuahua with a red cape, made of gold" so it understands which material to use. If you ignore the comma then only the cape will be made out of gold, so you have to be as clear as possible or it won't understand what you mean. If you only want the chihuahua to be in a specific material you type "superchihuahua made of gold with a red cape."
Glas
Endnote
The Bing Image Creator tool is supercool and it can generate really good images. The drawback comes when you want to make changes to the image, such as if you want another camera angle. How do you formulate that in text? Neither is the tool good at generating text. I tried to generate images with the text "YOLO" in them - but none of the images had that text in it - the closest image had the text YOLE. I also tried to generate images of a Swedish fighter jet called Jas 39 Gripen - and not a single one of them was good. You train these AI tools by showing them millions of images, so it's understandable it won't understand some niche product if not many images are available. But the tool is free and you get what you pay for.
I'm glad I'm not a full time artist because these tools will replace some of their work. They can still paint images with text in them or if you really need a specific image, such as a chihuahua from a certain camera angle or a product of which photos is not available in great abundance to train the AI.
”Monty Python”, “H R Giger”, “South park”, “M C Escher”, “Jackson pollack” are some other interesting styles to try.
ReplyDelete