You will also be interested
Three apps” text-to-image (AI image generator) have emerged as, a priori, the most interesting in the field.
- Dall.e 2, from OpenAI, which was the first to appear;
- Midjourney, often considered the best in terms of image quality, even if handling it is not easy. We have also dedicated a tutorial to it (here);
- Stable Diffusion, whose interface is simpler and which therefore makes it easy to obtain quality visuals.
Other applications have emerged, including these:
- Leonardo.ai which presents itself as a serious competitor to Midjourney;
- Maze Guru, very promising, but again not very easy to learn;
- Adobe Firefly, easy to use, with impressive visuals.
The concern with the aforementioned applications is that you usually have to register in order to request access, which is generally only granted after a few weeks.
From and 2
Dall.e 2 is the work of OpenAI, the creator of ChatGPT. However, it is not the most brilliant application in this area. However, it is likely that many users will discover the “ text to image with Dall.e 2 because this tool has been integrated into Microsoft’s Bing Image Creator, accessible from Bing. So you might as well know how to get Dall.e 2 to produce beautiful images. One of the keys will reside in the addition of a particular style. We will see many examples here.
Dall.e 2 is accessible from this address. Click on Try DALL-E to access the text typing interface.
We started from this request: a Blade Runner replica attends a social event in a retro-futuristic setting reminiscent of the 19th centurye century ».
As you can see if you try such a query yourself, the result is trivial. Also, quite often you will be able to notice imperfections in the designs.
Color, shadow and light
The first thing we can do is to add indications of color, shadow, light. We have also enriched the prompt so that it corresponds better to what we want: a Blade Runner replica attends a social event in a retro-futuristic setting reminiscent of the 19th centurye century, many extravagantly dressed guests, dim pastel colored lights ».
What to say ? We get a more promising but still far from satisfying delivery, compared to what we would get under Midjourney or Stable Diffusion.
Specify an art style
The key to a quality image in Dall.e 2 will consist in adding an image style.
In the style of a painter or graphic artist
A first option may be to request that the image be produced in the style of a famous painter or draftsman. Here is an example: a Blade Runner replica attends a social event in a retro-futuristic setting reminiscent of the 19th centurye century, many extravagantly dressed guests, Andy Warhol painting style “. To simplify, we have removed the indications of light and colors.
This artistic movement, which flourished during the 1920s, favored elegant geometric shapes. Let’s try this prompt: “ a Blade Runner replica attends a social event in a retro-futuristic setting reminiscent of the 19th centurye century, many extravagantly dressed guests, Art Deco style ».
Now, several of the images seemed more or less suitable for what we could be looking for.
The engraving technique leads to drawing lines in a metal plate to draw shapes. Asking Dall.e 2 to apply this style to the image, we get results that all turn out to be interesting. (From now on, we only replaced what was after “style” in the prompt.)
Vector art refers to art produced using vector illustration software (using mathematical formulas) such as Adobe Illustrator. Usually this style produces sharp and colorful images. This time we got several artistic quality images.
Vexel art is a digital art in which images are formed from matrices of colored dots, and used for some video games like Minecraft. It usually produces images rich in color and detail.
Ebru is a painting technique on the surface of water. It is an art born in Türkiye in the XVe century. When put to good use in Dall.e 2, the results are almost always of high quality.
And There you go. If an image particularly appeals to you, you can download it, knowing that the basic format is 1024 x 1024 px, which may seem insufficient if you want publishable material. So, if you want higher definition (and also vastly superior image quality), you might be better off switching to MidJourney or Stable Diffusion.
You can also try styles like Pop art, surrealism, Pop surrealism which is a combination of the two, claymotion (as in Wallace & Gromit) and any other art movement that appeals to you.
You can also take inspiration from images made by others; copy the corresponding prompts and adapt them to the image you want to obtain.
On the version accessible from the OpenAI site, you can load an image (upload image on the home page) and generate variations of it.