Big News: DALL-E 3 Now in Research Preview

DALL-E 3 Launch: ChatGPT Integration Enables Precise Text-to-Image Generation with Enhanced Safety Features

We are thrilled to announce that DALL-E 3, our latest AI image generation model, is now in research preview for select ChatGPT Plus and Enterprise customers! DALL-E 3 represents a major leap forward in creating images that precisely match text descriptions. Read on for all the exciting details.

The Power of DALL-E 3

DALL-E 3 builds off the groundbreaking work of DALL-E 2, but with significantly enhanced capabilities. While text-to-image models have historically struggled to fully capture nuanced prompts, DALL-E 3 was designed from the ground up to generate images that closely adhere to text.

Even when given the exact same prompt, DALL-E 3 produces noticeably sharper, more accurate images compared to DALL-E 2. Just take a look at this example prompt:

"An expressive oil painting of a basketball player dunking, depicted as an explosion of a nebula."

DALL-E 2 result:

DALL-E 3 result:

The DALL-E 3 image clearly depicts the "explosion of a nebula" described in the text, showcasing its improved fidelity.

Built on ChatGPT for Smarter Prompting

So how does DALL-E 3 achieve this boost in precision? It was built natively on top of ChatGPT, allowing the two models to work together seamlessly. This integration makes ChatGPT the perfect prompt brainstorming and refinement tool for DALL-E 3.

You can bounce ideas off of ChatGPT to come up with the perfect text description of the image you want to generate. ChatGPT will then automatically transform your description into a tailored, highly-detailed prompt formatted to get the best results from DALL-E 3.

If an initial DALL-E 3 image isn't quite right, just ask ChatGPT to tweak the prompt based on your feedback. After a few quick iterations, you'll have the exact image envisioned. It's an intuitive and powerful creative workflow made possible by the deep integration between these models.

Coming in October to Select Customers

We are beginning to roll out DALL-E 3 access in early October to ChatGPT Plus and Enterprise customers through the API and OpenAI Labs. As with DALL-E 2, you own the images you create with DALL-E 3 and can use them however you like, without needing permission.

We know many are eager to get their hands on this new creation tool, so stay tuned for wider availability soon!

Ongoing Focus on Safety

While rapidly advancing, AI image generation does come with risks around harmful content. As with all our models, safety is foundational to DALL-E 3's design.

Mitigating Potential Harms

We have implemented multiple strategies to mitigate potential harms:

- Blocking generation of images based on direct prompts for violence, adult content, and hate.

- Declining requests with public figures' names to limit impersonation risks.

- Partnering with outside experts to identify and address risks around issues like harmful biases and misinformation.

Testing for AI-Generated Content

We are also researching ways to help identify AI-created images. Our team has developed an internal "provenance classifier" that can effectively flag images generated by DALL-E 3.

By better detecting generated content, we aim to understand how these images might be misused, and explore solutions. Details on this tool are coming soon.

Upholding Creative Rights

Finally, DALL-E 3 has measures to protect creative rights. It declines creating images mimicking recognizable living artists' styles without permission. Users can also opt out of having DALL-E 3 trained on their original art.

The Future of AI Generation

DALL-E 3 represents a milestone in creating images that truly match imagined ideas. But there is still much work to be done in advancing and ensuring responsible use of this technology. As we continue developing DALL-E 3, safety remains our top priority. We aim to empower creativity, while mitigating harm.

We can't wait to see what you create with DALL-E 3! Be sure to share your amazing images and feedback as we expand access in the weeks ahead. The future of AI-enabled imagination is bright. Let's build it responsibly, together.