OpenAI's Dall-E 3 Now Supports Seed Number

It appears that OpenAI has quietly released a significant new feature for Dall-E 3—it now supports functional seed numbers!

What the heck does seed even mean?

Seeds act like a unique ID for each image Dall-E generates.

Seed numbers allow you to recreate the exact same image later. Dall-E 3 normally generates images randomly each time you provide a prompt. But with a seed number, you can force it to generate the same image again.

Previously, Dall-E 3 would always return a default seed number of 5000 for all the images it generated, which wasn’t helpful. Now it returns a large number.

Here’s an example:

Prompt: an image of a girl model holding a card that says “Dall-E is awesome”

As you can see, ChatGPT gave me the seed number 1458403096 for the first image.

Using that seed number, I can generate a similar image with minor tweaks by describing the changes I want.

I asked ChatGPT to change the text on the card from “Dall-E is awesome” to “I am cool”. In just seconds, it perfectly changed the text—very cool!

To test it further, I asked the AI to shorten the model’s hair. Again, it changed only the hair.

What about changing the model’s age—can it handle that?

Oh yes, it transformed the model into a young adult while keeping the clothes, pose, and card the same. Very impressive!

Next, I wanted a blue background. Usually you’d need Photoshop for this, but can Dall-E 3 do it?

It can! Although there are minor differences from the original, it still blows my mind that it worked.

Finally, I asked the AI to make the model pose sideways.

You could use this to easily generate images for ads or campaigns.

If you’re one who’s looking for models to promote your product, why pay when you can do it fast and cheaply with ChatGPT?

Why are seed numbers so important?

In the context of Dall-E 3 and image generation:

• Controlled Variability: By exposing the seed number, Dall-E 3 allows users to generate different variations of images based on the same prompt.
• Reproducibility: If you want to recreate a particular image or a set of images, you can do so by providing the same seed number that was used initially.
• Fine-tuning and Experimentation: You can fine-tune or experiment with the image generation process by tweaking the seed numbers. This feature can lead to the discovery of optimal seed numbers for particular kinds of image generation tasks.

Final Thoughts

Overall, I had so much fun doing experiments with seed numbers on Dall-E 3. The addition of a seed number feature in Dall-E 3 demonstrates a significant step towards giving users more control over the image generation process.

One thing to note, though: Dall-E 3 is not the first AI image generator to use seeds to control the image generation process. As far as I can remember, Stable Diffusion is the first AI image tool that allows the use of seeds in generating images.

‍