DALL-E is a generative artificial intelligence application developed by the American AI research laboratory OpenAI based on a deep learning model and natural language processing or NLP. It specifically generates images from natural language descriptions or prompts using Generative Pre-Trained Transformer 3 or GPT-3 autoregressive language model.
The name “DALL-E” is a portmanteau. OpenAI took inspiration from the animated robot character from Pixar WALL-E and acclaimed Spanish surrealist artist Salvador Dalì. It is part of the expanding list of generative AI applications and services from OpenAI alongside its ChatGPT chatbot and Whisper speech recognition model.
DALL-E 2 is the current version of the application which was introduced as a beta on 20 July 2022 to limited users and made available to the general public beginning on 28 September 2022. The application marks the advantages of deep learning and practical applications of NLP in the realms of generative AI and graphics design.
The Advantages of DALL-E: The Benefits of Using Artificial Intelligence to Generate Images
Central to the capabilities and advantages of DALL-E is that it can generate an image or a set of images in different styles such as photorealistic imageries, paintings, cartoons, and futuristic graphics through user-inputted text descriptions. A user can generate simple images from simple descriptions to more complex images from detailed descriptions.
At the heart of the application is the use of a large language model for deep learning training. The application can understand a particular text description as an instruction and produce intended outcomes. It is also based on a large collection of images that were collectively used as training data. The images it produces are a product of this dataset.
Below are its specific advantages:
• Digital Art Generation: The service is ideal for generating digital artwork. This can be useful for content creators such as bloggers and publishers, social media influencers, and social media and digital marketers, among others. The generated images do not have a license and can be used for commercial purposes.
• Editing and Manipulation: A user can also upload his or her chosen image and let DALL-E do certain image editing and manipulation capabilities through the use of prompts. A user-uploaded image, for example, can be transformed into different artistic styles such as oil painting or futuristic digital art.
• Ownership of Images: Remember that the generated images do not have a license. A particular user is free to do whatever he or she intends to do with these images. The actual copyright or other applicable ownership rights to these images are actually assigned to the person who generates them.
• Freemium Service: One of the biggest advantages of this application is that it is free. A user can generate 50 images for free in his or her first month. This translates to 50 free credits. The same user has 15 free credits on each succeeding month and he or she can purchase additional credits starting at USD 15.00 for 115 credits.
• Available as an API: OpenAI has released DALL-E as an API on November 2022. This has allowed other developers to integrate its core AI model into their own applications or services. Microsoft has integrated this API in its Designer app and Image Creator tool that is part of its Bing and Microsoft Edge products.
• Ethical Safeguards: There are concerns over using the service to propagate misinformation. These have been addressed through built-in safeguards that reject prompts involving public figures or uploads containing human faces. Prompts that have the potential to harm are prohibited and blocked.
The Disadvantages of DALL-E: Limitations and Issues in Using Generative AI to Produce Images
Remember that DALL-E is a specific application of generative AI. The outputs it produces are based on a collection of images that have been used as training data for the underlying deep learning model. It is possible for the application to produce somewhat similar images across different users and even across different generative AI platforms.
The main disadvantages of this application reflect the drawbacks and limitations of generative AI applications and other subfields of artificial intelligence and relevant AI concepts such as machine learning and deep learning models, natural language processing and large language modeling, and artificial neural networks.
Below are its specific disadvantages:
• Surreal Artworks: The application is more ideal for generating surreal artworks. It still struggles to produce photorealistic images and might not be ideal for producing certain types of images and other artworks such as a decent portrait of a person or realistic or natural imageries such as landscapes and cityscapes.
• Detailed Prompts: Another disadvantage of DALL-E is that its output or the image that it generates is dependent on the quality of a particular prompt. OpenAI recommends for a particular user to add as much detail as possible. There are times when the application still struggles to produce desired outputs.
• Language Limitations: It only accepts prompts written in the English language. Furthermore, despite its dependence on detailed prompts, its language processing has limits. It is unable to produce a desired outcome if a prompt contains more than three objects, negation, numbers, and connected sentences.
• Partiality Tendencies: The application also tends to generate a higher proportion of images of men than women for prompts that do not mention gender. This biased result comes from its reliance on public datasets. Note that one of the drawbacks of deep learning is that biased training data often produce biased outcomes.
• Possible Legal Issues: Using the generated image for commercial purposes may warrant possible legal issues such as violation of privacy rights or infringement of intellectual property rights. Remember that DALL-E uses a collection of images. These images might have limited rights while generate images can have similarities.
• Notable Criticisms: Generative AI applications have also been criticized for their potential to render certain professional creative jobs obsolete. Artists have also expressed concerns over the fact that these applications use the works of others that have been scrapped from a public database such as the internet.
