Initially just a fantasy, AI image generation has made impressive progress since its introduction a few years ago. Tools like Midjourney, DALL-E, and Jasper Art have proven to be incredibly powerful, allowing users to create almost anything they can imagine. OpenAI is now taking it a step further with the upcoming release of DALL-E 3, which will be available to the public later this year. So what exactly can we expect from DALL-E 3? Let’s compare it to DALL-E 2.
DALL-E 2 is effective at interpreting text prompts, although it does have some issues. It doesn’t always correctly understand all text prompts and struggles with generating certain objects. One of the major improvements in DALL-E 3 is its enhanced understanding of text prompts, especially longer ones. It has also addressed previous challenges in generating human details such as hands and reflections.
One new feature introduced in DALL-E 3 is the integration with ChatGPT. The AI chatbot has been built on top of DALL-E 3, allowing users to collaborate with it as a “brainstorming partner” for generating image ideas through conversational exchanges. Users who are new to AI image generation can utilize ChatGPT to iterate on their text prompts, with the AI assistant providing helpful suggestions to improve the image generation process.
When comparing the output of DALL-E 2 and DALL-E 3 for the same prompt, it’s evident that the new version produces significantly better images. DALL-E 3 generates images with greater detail, sharper lighting, more realistic textures, and more intricate backgrounds. In almost every aspect, the image quality surpasses that of DALL-E 2. Additionally, DALL-E 3 has improved text generation within images, which has historically been a challenge for AI image generation software.
While DALL-E 2 was only accessible through a standalone tool on OpenAI’s website, DALL-E 3 is now directly available through Microsoft’s search engine, Bing. Users can access the feature via Bing Chat and request prompts from the AI image generator. This integration was not available in DALL-E 2. However, it is exclusively available for ChatGPT Plus users who subscribe to a monthly plan.
Another significant change in DALL-E 3 is its focus on safety protocols. It prevents the generation of images featuring adult, violent, or hateful content. If a user submits a text prompt that requests the creation of inappropriate or explicit images, DALL-E 3’s safety protocols will flag the request and deny it. The safety measures also extend to avoiding copyright infringement by refusing to generate images resembling living public figures or imitating the styles of living artists. These changes have been implemented to comply with copyright laws and prevent the creation of offensive or harmful images.
DALL-E 3 was officially released in August 2023, but its widespread availability is currently limited to Microsoft’s Bing Chat. Researchers have access to Version 3.0, but it is not accessible to the general public in other forms.