The new neural network DALL-E 2 creates realistic images according to the description and generates similar images
Miscellaneous / / April 23, 2022
It's really impressive.
OpenAI introduced DALL-E 2, an updated version of the neural network, for the first time demonstrated in January 2021. Now it generates even better and more realistic images according to the description in English.
An example of generating for the query "Astronaut rides a horse in a photorealistic style":
1 / 0
2 / 0
3 / 0
4 / 0
5 / 0
6 / 0
7 / 0
8 / 0
In addition to generating from scratch, DALL-E 2 can make changes to existing images based on text corrections. She can add and remove elements, taking into account shadows, reflections and textures.
1 / 0
2 / 0
3 / 0
4 / 0
5 / 0
Another amazing ability of the new neural network is the creation of analogues of downloaded images (examples above). The neural network can literally redraw the picture in a different style, preserving all objects, or simply change the position of objects, their colors and proportions.
With all this, the resolution of the generated images has increased four times compared to DALL-E.
In OpenAI note, that while the project is only available for closed testing, but in the future the neural network will be opened to third-party developers who will be able to use all its features.
Read also🧐
- Sber launched the ruDALL-E neural network, which generates images according to the description