Reducing bias and improving safety in DALL·E 2
OpenAI announced a new technique applied to DALL·E 2 that adjusts image generation of people to better reflect global demographic diversity. The intervention targets representational bias in the model's outputs when generating human subjects. This is an early public example of a major lab deploying a post-training bias mitigation technique in a production image generation system.
Related guides (4)
Related events (8)
DALL·E 2 Pre-Training Mitigations
OpenAI describes the safety mitigations applied during pre-training of DALL·E 2 to reduce risks from powerful image generation models. The post outlines guardrails designed to prevent generated images from violating content policy. This represents an early public disclosure of safety-by-design approaches for large generative image models.
DALL·E 2: Extending Creativity — Early Access Artist Program Update
OpenAI reports that over 3,000 artists from 118+ countries have joined the DALL·E 2 research preview and integrated the model into their creative workflows. These early access participants have informed feature decisions and surfaced new use cases for the image generation system. The update signals OpenAI's strategy of using artist communities as both testers and stakeholders in shaping generative image tools.
DALL·E 2 Research Preview Update: Expanding Access to 1,000 Users Per Week
OpenAI announced an expansion of the DALL·E 2 research preview, moving from a limited early-access phase to adding up to 1,000 new users per week from its waitlist. Early users have collectively generated over 3 million images, and their usage has informed improvements to OpenAI's safety processes. The update signals a gradual scaling of access rather than a full public launch.
DALL·E 3 System Card
OpenAI published the system card for DALL·E 3, its latest text-to-image model. The document covers safety evaluations, mitigation strategies, and known limitations related to harmful content generation, bias, and misuse risks. It follows OpenAI's practice of releasing system cards alongside major model deployments to document safety testing and residual risks.
DALL·E API now available in public beta
OpenAI has opened public beta access to the DALL·E API, enabling developers to integrate image generation capabilities into their own applications. This marks the transition of DALL·E from a standalone product to a platform-accessible service. The announcement expands the ecosystem of AI-powered image generation tooling.
DALL·E 3 Now Available in ChatGPT Plus and Enterprise
OpenAI has rolled out DALL·E 3 to ChatGPT Plus and Enterprise subscribers, expanding access to its latest image generation model. The announcement also highlights a safety mitigation stack developed for the wider release and provides updates on provenance research aimed at identifying AI-generated images.
DALL·E: Introducing Outpainting
OpenAI announced outpainting for DALL·E, a capability that allows users to extend images beyond their original borders to any size. The feature uses DALL·E's generative capabilities to fill in new regions consistent with the existing image content. This expands creative use cases for the model by enabling larger compositions and narrative storytelling through image extension.
DALL·E Now Available in Beta
OpenAI is opening DALL·E to 1 million waitlisted users in a public beta. Users receive free monthly credits and can purchase additional credits in 115-generation increments for $15. This marks the first broad public access to DALL·E's text-to-image generation capabilities.