OpenAI widens DALL·E 2 access and adds safety mitigations

OpenAI said it would begin inviting roughly a million people off the DALL·E 2 waitlist over the following weeks, expanding a preview that had, since April, been restricted to a few hundred vetted testers. Signups had opened without an invitation requirement, and the company framed the move as the next stage of a gradual, safety-monitored rollout rather than a full public launch.

The expansion came with product and policy changes alongside the raw increase in scale. OpenAI added content filters intended to catch a wider range of policy-violating prompts and outputs than the original preview’s filtering had covered, and it began attaching a visible signature to every generated image to make AI-generated origin identifiable — a provenance measure aimed at the disinformation risk the April system card had flagged. The company also allowed users to upload photos of real faces for editing for the first time, a capability it had withheld at the smaller preview stage over harassment concerns, now paired with additional restrictions on that specific feature.

The access expansion also marked DALL·E 2’s shift from a free research preview toward a commercial product: within weeks, OpenAI introduced a credit-based pricing structure, giving new users a batch of free monthly credits and charging for additional generations thereafter. Going from a few hundred users to roughly a million in the space of months made DALL·E 2 the first widely-used consumer image generator from a major lab, arriving shortly before Stability AI’s open-weight Stable Diffusion changed the competitive picture entirely.