• ThePrompt
  • Posts
  • Analysis: Dalle-3 VS MidJourney

Analysis: Dalle-3 VS MidJourney

As we all know it, Dalle-3 is coming to ChatGPT+ soon. It's a big deal because MidJourney is currently the top dog for making images.

Both have cool “strenghts” to make even better images.

We're here to help you understand what's going on so you can be smarter about AI and images 👇🏻

What are the strengths of Dalle-3?

  1. Talking helps make it better: When people chat and say what's wrong with a picture, it's great info or human feedback for training. Integrating Dalle-3 into ChatGPT is a genius move, because it will help both the language and the image parts of GPT-4 get better and much faster. This is the most important part of the puzzle.

  2. Better tech, less data: MidJourney has lots of data but isn't as good. OpenAI uses smarter tech (e.g. Consistency Model), that learns better even with less data — if this works well with Dalle-3 they can improve at a much faster rate than MJ.

  3. Everything fits together: Combining ChatGPT with Dalle-3 makes everything better. You can easily add more stuff like code understanding or image searching. And users will only need to pay one subscription to use both text and images with OpenAI, so MJ adoption may stall.

  4. More people, better reach: ChatGPT has way more users than MidJourney. It's easier to get it out to people, and it's not stuck in places like Discord which can be hard for new users.

These all sound nice, but Midjourney has been this “weird thing” that works perfectly well, and no-one knows how they pull it out.

Maybe they have more surprises. Elon Musk said it himself 👇🏻

So, what are the strengths of MidJourney?

You can only use MJ in Discord, but their UI is a “hidden-gem” for model improvement and usage growth.

MidJourney makes it very easy for you to tell them if you like a photo. As you use the product, you’re also helping it get better.

If you pick or “upscale” your favorite image out of the 4 choices, you're already helping the system learn what's good. No need to do anything extra, just use it and you're helping out.

In this department, OpenAI has a big problem — They have to pay people to say what's good or bad because users of ChatGPT don't have good incentives to help out.

The 👍🏻👎🏻 may be clicked 2.5% of the time, and that’s not enough value to get the average of good answers over all user base.

And that’s the quick analysis, hope you liked it!

Get AI smart in 3 minutes or less

Subscribe to our AI newsletter “The Prompt” where we unpack the latest news, trends and tools as simple as chatting with our friends.