I'm still using Bing's image generator every day, or just about (today I embedded YouTube videos instead, but yesterday I definitely used it). I'll be a keen observer of what sorts of "improvements" have been made.

One thing I noticed was that Bing (Dall E) seems to be really slow right now. I also hit a snag asking for something in the style of a particular artist; I'm wondering if one improvement was more on the legal front.

Expand full comment

The "style of a particular artist" is definitely by design. I actually mentioned it in my last week's post about DALL-E 3. Here's exactly what their "A focus on safety" section says about it:

"DALL·E 3 is designed to decline requests that ask for an image in the style of a living artist. Creators can now also opt their images out from training of our future image generation models." (https://openai.com/dall-e-3)

As for improvements to prompt adherence, they're striking! Check out the example from my Notes post earlier today: https://substack.com/@whytryai/note/c-41131892

So DALL-E 3 can now:

1. Render legible text (it's hit and miss and may take a few rerolls)

2. Is much, much better at following descriptions

But the downside could be the "safety" features that make mimicking a living artist or specific person more difficult.

Won't be surprised if it's especially slow because of all the people rushing to test out free DALL-E 3 generations.

Expand full comment

It's better and worse at the same time! Better and worse at the same time is the new better.

Expand full comment

Honestly, I welcome any steps to proactively solve the ethical and legal issues that most AI tools have been accused of (stealing styles wholesale, etc.).

You should still be able to get a visual close to the one you want by describing the style instead of mentioning the artist's name directly.

We'll probably see regulations that will impose this type of approach in general, so I'm okay with OpenAI doing this voluntarily in advance.

To me, being able to describe a scene using natural language and have AI actually follow it precisely is a major step in the direction of opening these tools up to a broader audience.

Expand full comment

I'm not being critical of their approach in any way, just to be clear.

We are going to have plenty of backs and forths like this, as society adapts to technology. It's a classic struggle we get to see unfold again, which is simultaneously exhilarating and frustrating.

Expand full comment