DALL·E: Creating Images from Text

The current advancement in neural networks lets to perform text generation or image generation tasks. Researchers have recently extended such abilities and introduced a neural network DALL·E that can convert text descriptions to images.

The system can use the compositional structure of language to generate plausible images for lots of sentences. For instance, it lets to render the image from the desired viewpoint or according to the chosen 3D style (as visualize a turtle made of wireframe). It is possible to depict the internal and external structure by cross-sectional views or macro photographs.

Furthermore, DALL·E can combine unrelated concepts to produce imaginary views as giraffe turtle chimera. It can also use a cue in the description to generate the image, for instance, visualize a “photo reflected upside-down.” The study shows that current technologies can manipulate visual concepts through language.

Link to the page of project: https://openai.com/blog/dall-e/