I have been busy dealing with some trivial matters recently, which has affected the progress of AI learning. Now I will record my current thoughts.
I am trying to use chatGPT and Midjourney as AI assistants to achieve the effect of one person being able to become an army. I focus on thinking about the core content and let AI assist in implementation to see if it can improve work efficiency.
I have been using Midjourney (MJ) for a while now. The problem I am currently facing is that the generated images are quite different from what I have in mind. If it's just a simple idea, like "a man standing on a mountain looking into the distance," MJ can generate many different styles of images, which helps expand my thinking. However, when the demand for details increases, the situation becomes difficult to control. For example, when I add the detail of "holding a robe in hand," MJ has a hard time accurately presenting this scene. It either puts the robe on the person or introduces a character related to the robe. In other words, if the precision of expressing the content in the image is not high, MJ's generation effect is quite good. But if higher accuracy is required, it is necessary to improve the accuracy through continuous prompts. I still need to delve into this aspect of learning. Fortunately, I found a good learning resource that can be referred to: Midjourney Learning Path.
Next, I plan to study SD and compare it with MJ to understand the advantages and disadvantages of each.