AI Pulse by Inblix

OpenAI's DALL·E 2 hits 3M images, opens to 1,000 new users weekly

OpenAI Blog · Jul 18, 2026 · 2 min read · Read original article →

Curated by the Inblix editorial team


Featured image for article: OpenAI's DALL·E 2 hits 3M images, opens to 1,000 new users weekly

Three million images in a month. That’s the collective output from early testers of OpenAI’s DALL·E 2, a figure the company dropped today alongside a significant update: the waitlist is finally moving. Starting this week, up to 1,000 new users will be onboarded every seven days as the research preview shifts into a slightly wider release.

OpenAI framed the expansion as a direct result of real-world feedback. Since the initial preview launched for a hand-picked group of “trusted users” last month, the company has been quietly overhauling its safety infrastructure. Text filters got a boost, and the automated detection systems that flag content policy violations have been tuned based on actual usage patterns. The numbers they shared tell an interesting story—less than 0.05% of downloaded or publicly shared images tripped the filters, and of those, only about 30% were confirmed violations by human reviewers. That’s a tiny fraction of the total, though even a small percentage at three million images translates to real moderation work.

The company is also grappling with the thornier issue of bias baked into the model’s training data. Their stopgap approach is pragmatic rather than technical: early users were told not to share photorealistic images containing faces, and to actively flag problematic outputs. OpenAI says they believe this blunt instrument has been “effective in limiting potential harm” and will keep the policy in place for now. It’s a reminder that governance here is still a work in progress—don’t expect unfiltered photorealism anytime soon.

What’s clear is that the slow drip of access is deliberate. OpenAI admits they want to build confidence in their safety systems before speeding up onboarding. For the thousands still languishing on the waitlist, 1,000 invites a week feels like a trickle, but the company’s logic is consistent: better to measure twice and cut once when the tool generates anything you can describe in a sentence. In the meantime, they’re pointing the curious toward their Instagram account, @openaidalle, for a taste of what people have already dreamed up.

💡 Key Takeaways

  1. OpenAI's DALL·E 2 users have generated over 3 million images during the research preview, a scale that has directly informed the company's safety system improvements.
  2. Only 0.05% of downloaded or shared images were flagged as potentially violating content policy, and just 30% of those were confirmed violations by human reviewers.
  3. To mitigate training data bias, OpenAI has restricted sharing of photorealistic face generations and asked users to report problematic outputs rather than relying solely on automated fixes.
  4. The company will onboard 1,000 new users per week as it iterates on safety, with no firm timeline for a broader public launch.

Keep reading: See related articles below for more coverage on this topic.

Get smarter about AI

The sharpest AI news, curated daily. Delivered free to your inbox.

Learn more

Glossary terms

← Back to all articles