Is DALL-E 2 Real?

In January 2021, OpenAI (a research group funded by Elon Musk, Microsoft, and other Silicon Valley visionaries) launched DALL-E, a system that uses artificial intelligence (AI) to generate images from text prompts and source images. A year later, the group launched DALL-E 2, which is an improved version of the first system. According to OpenAI, DALL-E 2 can perform more tasks than DALL-E.

However, some people are skeptical about the existence of any AI art generator that can do all the incredible things that DALL-E 2 is said to be capable of. Many of the people wanting to try DALL-E online are asking: Is DALL-E 2 real and who can use DALL-E 2?

What Is DALL-E 2?

DALL-E 2 is a new version of the original DALL-E. The two programs are AI systems created by OpenAI to enable professional and amateur artists to create digital images from text prompts and source images. The DALL-E system uses a 12-billion parameter version of the Generative Pre-Trained Transformer (GPT-3) model to convert text inputs into corresponding images.

OpenAI developed the GPT-1 Transformer model in 2018 based on the Transformer architecture. A year later, it was upgraded to GPT-2. The GPT-3 model was developed in 2020 and features 175 billion parameters. The model utilised by DALL-E is a multimodal application of GPT-3 with 12 billion parameters. This means it can easily exchange text prompts for pixels.

What Does DALL-E 2 Do?

DALL-E 2 is an advanced version of its predecessor and can generate realistic images and pieces of art from simple descriptions in natural languages. For instance, if you want to generate an image of an astronaut riding a white horse in space, you simply input these words into the user interface and DALL-E 2’s algorithm will do the rest. This system merges concepts, features, and styles to generate original, realistic digital images and art.

It also allows you to edit existing images from text descriptions in natural languages. These edits are precise and realistic to give you the kind of image you want. Furthermore, you can add or remove certain components of the existing image without affecting important features such as shadows, textures, and reflections. With DALL-E 2, you can reproduce an image to generate different variations inspired by the original piece.

This version of DALL-E also understands the connection between images and texts, describing them through a process referred to as diffusion. DALL-E 2 starts by creating a pattern of disorganised dots and eventually adjusts the pattern to form the right image once it fully recognises the exact features of the image.

What Makes DALL-E 2 Better Than DALL-E?

As noted above, DALL-E 2 is an upgraded version of DALL-E and offers a better user experience than its predecessor. For instance, DALL-E 2 is reported to be able to generate images that are more accurate and realistic than the previous version. Also, images generated by DALL-E 2 have four times superior resolution.

Another reason why users prefer DALL-E 2 over the previous version is its ability to match photo descriptions and photorealism. However, both versions are still not available to the public. According to OpenAI, both DALL-E  and DALL-E 2 aren’t available in its application programming interface because the company is committed to ensuring that its AI art generators are developed and deployed responsibly. So far, both art generation tools have only been available to people lucky enough to make it onto the waiting list.

The limited use of DALL-E 2 and DALL-E has helped the organisation regulate the types of products being created and published on the Internet. Even the privileged users are prohibited from generating images or pieces of art that contain content related to violence, nudity, sex, hate, or harassment. 

Currently, DALL-E 2 is only available to a few hundred users around the world. These users were carefully chosen to help OpenAI carry out its beta testing. In May 2022, the number of DALL-E 2 users was only 640. The select group of users helps the organisation identify areas that need to be corrected before the system is released to the public.