DALL-E 2 is a text-to-image model developed by OpenAI, announced on 6 April 2022 as the successor to the original DALL-E. It uses deep learning methodologies to generate digital images from natural language descriptions, known as prompts. The model was designed to produce more realistic images at higher resolutions than its predecessor, with the ability to combine concepts, attributes, and styles. Its name is a portmanteau of the Pixar robot character WALL-E and the Spanish surrealist artist Salvador DalÃ.
History and background
The original DALL-E was revealed by OpenAI in a blog post on 5 January 2021, using a modified version of GPT-3 to generate images. DALL-E 2 was announced on 6 April 2022, entering a beta phase on 20 July 2022 with invitations sent to 1 million waitlisted individuals. During this phase, users could generate a certain number of images for free each month and purchase more. Access had previously been restricted to pre-selected users for a research preview due to concerns about ethics and safety. On 28 September 2022, DALL-E 2 was opened to everyone, and the waitlist requirement was removed. In early November 2022, OpenAI released DALL-E 2 as an API, allowing developers to integrate the model into their own applications. Microsoft subsequently implemented DALL-E 2 in its Designer app and Image Creator tool included in Bing and Microsoft Edge. The API operates on a cost-per-image basis, with prices varying by image resolution and volume discounts available for enterprise customers. In February 2024, OpenAI began adding watermarks to DALL-E generated images, containing metadata in the C2PA (Coalition for Content Provenance and Authenticity) standard promoted by the Content Authenticity Initiative.
Technology
DALL-E 2 uses 3.5 billion parameters, a smaller number than its predecessor. Instead of an autoregressive Transformer, it employs a diffusion model conditioned on CLIP image embeddings. During inference, these image embeddings are generated from CLIP text embeddings by a prior model. This architecture is the same as that of Stable Diffusion, released a few months later. The model builds on earlier work in Generative AI and Deep learning, particularly the development of the Transformer (architecture) architecture by OpenAI and others. The first generative pre-trained transformer (GPT) model was developed by OpenAI in 2018, scaled up to GPT-2 in 2019 and GPT-3 in 2020, with 175 billion parameters. DALL-E 2's diffusion approach differs from the autoregressive method of the original DALL-E, which used a discrete VAE and a decoder-only Transformer model with 12 billion parameters.
Capabilities
DALL-E 2 can generate imagery in multiple styles, including photorealistic imagery, paintings, and emoji. It can manipulate and rearrange objects in its images and correctly place design elements in novel compositions without explicit instruction. The model exhibits a broad understanding of visual and design trends, and can produce images for a wide variety of arbitrary descriptions from various viewpoints with only rare failures. Its visual reasoning ability is sufficient to solve Raven's Matrices, visual tests often administered to humans to measure intelligence. DALL-E 2 also supports image modification, including generating variations of existing images and editing them through inpainting and outpainting. These abilities use context from an image to fill in missing areas using a medium consistent with the original, following a given prompt. For example, outpainting can expand an image beyond its original borders, taking into account existing visual elements such as shadows, reflections, and textures.
Impact and legacy
DALL-E 2 marked a significant advancement in text-to-image generation, bringing the technology to a broader public through its open beta and eventual removal of the waitlist. Its release spurred interest in Generative AI and influenced subsequent developments in the field, including DALL-E 3, released in October 2023 natively into ChatGPT for Plus and Enterprise customers. DALL-E 3 was later integrated into Microsoft's Bing Image Creator and Copilot, and in March 2025 was replaced in ChatGPT by GPT Image's native image-generation capabilities. The model's diffusion-based architecture also influenced other systems, such as Stable Diffusion, and contributed to the rapid evolution of Machine learning applications for creative tasks. DALL-E 2's approach to safety, including restricted access during research preview and later watermarking, set precedents for responsible deployment of generative models.
Reception and concerns
The release of DALL-E 2 generated considerable public and academic attention, highlighting both the creative potential and ethical challenges of AI-generated imagery. Concerns included potential misuse for creating misleading content, copyright issues, and the impact on artists and designers. OpenAI's phased rollout, with initial restrictions and a waitlist, reflected these concerns. The model's ability to generate realistic images from text prompts also raised questions about provenance and authenticity, leading to the adoption of C2PA watermarks in 2024. Despite these challenges, DALL-E 2 was widely praised for its technical achievements and user-friendly interface, helping to popularize text-to-image generation among non-specialists.