Midjourney is one of the most successful AI image generator companies. You write a prompt of what you want to create, and Midjourney generates an image for you. This prompt-to-generation model (also used to generate text with ChatGPT and others) implicates a new UX model with users, mainly because to develop the correct result, you typically need to try and refine your prompt a few times (e.g., prompt engineering).
Ideally, you want the model to run as fast as possible, but there is a tradeoff between the quality of the image created and how fast it runs. A smaller model will run faster but will yield a lower-quality output.
For Midjourney, 60 seconds ended up being the sweet spot. The initial model took 20 minutes and was too long to provide re-engagement. They tried a 10 seconds model, but it needed to provide better quality. The 60 seconds threshold proved to be the perfect balance of users willing to wait, see the results and refine it while being happy with the overall quality.
It is interesting to think about what potential implications in the future. For example, is there a world where you have two models, one for prototyping and one to generate the real version? Or will prompt definition be mostly abstracted from users by specialized vertical vendors (e.g., Lensa avatar AI)?
Craving more? Check out the source behind this Brain Snack!
