News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
imgDALL-E

DALL-E

Text-to-image model series by OpenAI

OpenAI developed this series of text-to-image models to create digital pictures from natural language prompts. The first version launched in 2021, followed by improved iterations that integrate with chat interfaces and enterprise tools.

Share on WhatsApp

  • First version announced in January 2021 by OpenAI
  • DALL-E 2 uses diffusion architecture with 3.5 billion parameters
  • Integrated natively into ChatGPT Plus in October 2023
  • Replaced by GPT Image in ChatGPT in March 2025
  • Adds C2PA watermarks to generated images since 2024

What it is

DALL-E is a family of artificial intelligence models built by OpenAI. It converts written descriptions into digital images using deep learning techniques. The initial version appeared in January 2021, utilizing a modified GPT-3 architecture. Subsequent releases, including DALL-E 2 and DALL-E 3, improved resolution and prompt adherence. The system operates as a proprietary service rather than open software. It serves creative professionals, developers, and general users through various platforms. The name combines references to the robot WALL-E and artist Salvador Dalí.

How it came about

OpenAI revealed the first model on 5 January 2021 via a blog post. The successor, DALL-E 2, arrived in April 2022 with higher resolution capabilities. A beta phase began in July 2022 for waitlisted users before opening to the public in September. DALL-E 3 launched in October 2023, integrating natively with ChatGPT Plus and Enterprise tiers. Microsoft incorporated the technology into Bing Image Creator and Designer apps. In March 2025, GPT Image replaced DALL-E 3 within the ChatGPT interface for native generation tasks.

Inside the model

The original model uses three parts: a discrete variational autoencoder, an autoregressive Transformer with 12 billion parameters, and CLIP encoders. It processes tokenised captions and image patches. DALL-E 2 shifted to a diffusion architecture conditioned on CLIP embeddings, using 3.5 billion parameters. This design mirrors Stable Diffusion. DALL-E 3 focuses on prompt following nuances, though technical training details remain undisclosed. Watermarks using C2PA standards were added in February 2024 to track provenance.

What it can do

The system generates photorealistic images, paintings, and emojis from text. It can rearrange objects and infer missing details like shadows or holiday themes. DALL-E 3 handles complex prompts with greater accuracy and renders coherent text better than prior versions. Users can edit existing images through inpainting and outpainting features. However, it struggles with negation, specific numbers, and distinguishing similar object attributes. Scientific imagery and typography often appear incorrect or gibberish.

How to use it

Access is provided through proprietary APIs and web interfaces. DALL-E 2 became available via API in late 2022, charging per image based on resolution. DALL-E 3 integrates into ChatGPT subscriptions and Microsoft Copilot tools. Volume discounts exist for enterprise clients. The weights and source code are not public, distinguishing it from open-source alternatives like Stable Diffusion. Users must adhere to usage policies that restrict certain content types.

Safety and testing

Training data underwent filtering to remove violent or sexual content, which inadvertently increased gender bias. The system inserts invisible phrases to balance demographic representation. It blocks prompts involving public figures or offensive material. Critics note that filters can be bypassed with alternative wording. In January 2024, OpenAI removed bans on military use cases. The model refuses to mimic living artists' styles to address copyright concerns.

Reception and debate

Media coverage often highlights surreal outputs like radishes in tutus. Investors viewed the technology as a potential industry turning point, contributing to significant funding rounds. Artists have criticized the tool for using scraped work without permission and undermining human intent. Some users complain about excessive content filtering that limits creative utility. Legal questions regarding image ownership remain unresolved despite OpenAI's permission grants for commercial use.

Why it matters

The release spurred development of open-source competitors such as Craiyon and Stable Diffusion. It demonstrated that language models could grasp everyday visual concepts effectively. The integration into mainstream tools like Bing and ChatGPT accelerated adoption of generative AI. Debates over copyright and artistic labor intensified following its popularity. The shift toward diffusion models in DALL-E 2 influenced broader architectural trends in the field.

Questions people ask about DALL-E

What is the difference between DALL-E 2 and DALL-E 3?

DALL-E 2 uses a diffusion model with 3.5 billion parameters for realistic images. DALL-E 3 focuses on understanding nuanced prompts and generating accurate text within images, integrating directly with ChatGPT.

Can I use DALL-E images for commercial purposes?

Yes, OpenAI states that users have permission to reprint, sell, or merchandise images produced by the models. However, legal concerns regarding copyright of training data remain inconclusive.

Why does DALL-E sometimes fail with simple prompts?

The model has limitations with negation, specific numbers, and connected sentences. It may confuse object attributes, such as placing colors on the wrong items, or struggle with scientific accuracy.

Good reading and sources

Where this page comes from

The facts on this page come from the Wikipedia article "DALL-E", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryLLM rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.

It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

The weekly digest

Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.