Do you still remember the 12 billion parameter magic model DALL·E released by OpenAI in January last year?
At that time, DALL·E’s style of painting was like this:
if only“read” text, DALL·E can “automatically” generate a lifelike master portrait based on the content of the text. Therefore, as soon as it was released at that time, DALL·E became popular in the AI circle, attracting countless fans, and it also made the Language-Vision (text-vision) direction popular again.
Just recently, a year later, OpenAI combined CLIP and released the second version of DALL·E——DALL・E 2.0!
DALL·E 2.0 can generate more realistic and accurate portraits than DALL·E 1.0: given in the comprehensive text descriptionConcept, Properties and StyleWait for three elements to generate “realistic” images and artworks!The resolution has been increased by 4 times!
For example, when the prompt text contains “concept” “An astronaut” (an astronaut), “property” “riding a horse” (riding a horse) and “style” “in a phtprealistic style” (surreal style) )Time:
Text prompt: An astronaut + riding a horse + in a phtprealistic style
DALL・E 2 can generate the followingContains three elements at the same timeImage:
On the basis of satisfying the three given elements, it gives full play to its “imagination”. Not only does the horse have different postures, but also puts on different styles of clothing for the astronauts. The scenes are also very rich. On the top of the mountain, on the top of the mountain, in the starry sky…
God is not miraculous! Bull is not bull!
Let’s enjoy the masterpieces of DALL・2.0!
If you want to transform one or more of the three elements of concept, attribute and style in the text, for example, keep “concept” “an astronaut” and “attribute” “riding a horse”, and change the surreal style Instead of pop artist Andy Warhol’s style, DALL·E 2 can also “easily” convert its painting style:
Text prompt: An astronaut + riding a horse + in the style of Andy Warhol
Image generated by DALL・E 2:
Text hint: An astronaut + riding a horse + as a pencil drawing
Image generated by DALL・E 2:
Text prompt: An astronaut + lounging in a tropical resort in space + in a vaporwave style
Image generated by DALL・E 2:
Text prompt: Teddy bears + mixing sparkling chemicals as mad scientists + as a 1990s Saturday morning cartoon style)
Image generated by DALL・E 2:
Text prompt: Teddy bears+shopping for groceries+in the style of ukiyo-e
Image generated by DALL・E 2:
Text prompt: Teddy bears+shopping for groceries+in ancient Egypt
Image generated by DALL・E 2:
Text prompt: A bowl of soup+that is a portal to another dimension+as digital art
Image generated by DALL・E 2:
Text prompt: A bowl of soup+as a planet in the universe+as a 1960s poster
Image generated by DALL・E 2:
Text prompt: A bowl of soup+as a planet in the universe+as digital art
Image generated by DALL・E 2:
Other Features of DALL・E 2.0
1. Image editing
DALL·E 2 enables realistic editing of existing images based on titles described in natural language, for example, adding or removing an element from the image while taking into account shadows, reflections and textures. An example is as follows:
Editing requirements for the text description: Choose a location to add the flamingo to the diagram.
Original image vs. DALL・E 2 edited image:
Editing requirements for text descriptions: Choose a location to add corgis to the diagram.
Original image vs. DALL・E 2 edited image:
2. Style Variations
DALL・E 2 can take a picture and then create different portraits of the same style based on the original picture. An example is as follows:
Original image 1:
Image of the same style created by DALL・E 2:
Original image 2:
Image of the same style created by DALL・E 2:
Original image 3:
Image of the same style created by DALL・E 2:
Original image 4:
Image of the same style created by DALL・E 2:
For the images generated by AI from text, we are of course the resolution of the image. The higher the resolution of the image, the greater the number of pixels, and the clearer and more realistic the image will be. Compared with DALL・E 1,DALL・E 2 has a 4x increase in resolution!
For example, for the same text prompt:
Text prompt: a painting of a fox sitting in a field at sunrise in the style of Claude Monet (a fox sitting in a field at sunrise + Claude Monet style)
The following two figures are a comparison of the images generated by DALL·E 1 and DALL·E 2:
In contrast, the image generated by DALL·E 1 can be said to be very blurred, and you can’t even see where the “sunrise” is at all, the “fox” only shows its head, and the “field” is not too fieldy. It looks like, and is far from the Impressionist style of the painter Monet in the overall style.
And under the magic of the DALL·E 2, the image quality is significantly improved, “Sunrise” and “Field” are very vivid, and the little fox is sitting on the grass with a cute posture. The painting is richer in color, uses more complex colors, and depicts light and shadow closer to Monet’s style.
Overall, compared to DALL·E 1.0, DALL·2 can obviously hold images with richer elements and fuller colors. It is no longer a simple description of a single item, but an overall expression of a scene, with a more complete story and richer imagination!
For more details, you can check the related research papers of DALL・E 2:
Paper address:https://cdn.openai.com/ papers/dall-e-2.pdf
.
[related_posts_by_tax taxonomies=”post_tag”]
The post Given 3 words, AI draws directly! OpenAI releases DALL·E 2.0, mastering a variety of painting styles, and the resolution is increased by 4 times appeared first on Gamingsym.