You’ve tried? DALL·E 2 is the newest artificial intelligence model from OpenAI. If you’ve seen some of his creations and think they’re amazing, you’re absolutely right.
OpenAI published an entry on their official blog and a white paper titled “Generating Hierarchical Text Conditional Images with CLIP Latents” on DALL E 2. The white paper is great for understanding the technical details but none of them explains how awesome DALL E 2 is, and the not so amazing, in depth.
Despite its size, DALL·E 2 generates images with four times better resolution than DALL·E and is preferred by human judges 70% of the time, both in subtitle matching and in photorealism.
DALL-E 2 can even modify existing images, create image variations that maintain their salient features, and interpolate between two input images. The impressive results of the DALL-E 2 have many wondering exactly how such a powerful model works under the hood.
How do you generate images?
While the DALL-E 2 can perform a variety of tasks, including image manipulation and interpolation as mentioned above, in this article we will focus on the imaging task and its echo in the real world.
The mental representation encodes the main features that are semantically significant : people, animals, objects, style, colors, background, etc. so that DALL·E 2 can generate a novel image that retains these features while varying non-essential features.
DALL-E 2 represents a step change in AI imaging technology . It understands natural language cues better than anything before it, allowing an unprecedented level of control over themes, styles, techniques, angles, backgrounds, locations, actions, attributes, and concepts, and generates images of extraordinary quality.
With high quality, it generates dozens of options in seconds, each at a level of quality that would take a human photographer, painter, digital artist, or illustrator hours to produce. It’s kind of an art director’s dream.
OpenAI explicitly says that users “get all rights to commercialize the images they create with DALL-E, including the right to reprint, sell, and trade.” But there are still some legal gray areas that have not yet been fully explored.
How long will it be before we see a complete short film, written, directed, soundtracked and made entirely by AI systems? What will the future be like then for artists?
Reference: OPENAI PROJECT
