ChatGPT is the newest talk of the town in AI. The old Version 3 was neat, but the latest 3.5 is just so smart that it’s creepy. GPT, or Generative Pre-trained Transformer, is a text-to-text AI that attempts to respond as if you are having a conversation with another person. It can be used to chat with, write stories, poems, or even code a script for Photoshop. Yes, seriously. Like other AI models, it is trained and cannot take in more information, however, during a single chat session it can remember what you tell it.
Months ago I used the old version to assist with MidJourney prompt writing, but I thought if the new version can remember things I tell it, maybe I can teach it how MidJourney works? I then wrote out how prompting works, format, algorithms, aspect ratios, and even what parameter settings work with different versions… and you know what? It worked!
All I do is copy and paste my “lesson”, hit the voice-to-text button and speak what kind of prompt I am looking for. After it gives you a result you can even ask it to replace things, change the version, or alter the prompt text. I foresee prompting changing in many ways, but one clear need is a voice-to-image interface - and something like this might be the way… Now, it’s not perfect. It has accidentally given me --stylize 5000 on --v 4 which isn’t allowed, but this is just a small test, and I bet if GPT gets to version 4 or 5 and can be actually trained on all of MidJourney - we will have an app on allowing us to make images by just talking to our phone.
(click the prompt to copy it!)
/imagine prompt: a futuristic city once bustling with life is now being overtaken by nature. Vines and plants grow through the cracks in the concrete, and trees tower above the crumbling skyscrapers. The setting sun casts a warm glow over the scene, creating a peaceful contrast to the abandoned city. --v 4 --style 4a --chaos 20 --stylize 70 --seed 420 --ar 3:2
(click to view)