Collection of GPT-image-generator 2.0 issues, bugs, and work-around tips (check first post)

Only some small tips. Every time a model is upgraded, even within the same version number, the style can change. Unfortunately you then no longer have access to the old models. What I have done for some work is to create a list of instructions and use it in a MyGPT. However, that is limited to 8k characters. But it is usually enough for tasks like this. You can first create a small protocol with short, concise instructions for GPT on how it should behave in this session. Then you start this session with that protocol. I assume you have been using the memories in the chat system so far. They are limited and less reliable and harder to maintain. An instruction protocol is better suited for this. Once the protocol works well, you turn it into a MyGPT and always start your session with it when you create images, and name it Narthmor. And if you use several styles, use several MyGPTs as well. Then the MyGPT would be more responsible for the overall style, and a small start protocol for fine adjustments. To become a good prompter, you do not necessarily have to understand the system technically. It only helps with intensive testing. Intuition for what exactly the words you use trigger, and imagination are much more important. And of course everyone can post here what they have found out. It is called bugs and issues. Bugs are more technical. Issues are more about experiences with prompts. It does not necessarily mean that only problems have to be discussed here. What ever tricks or experiences someone want share is welcome here. It just should not be a classic gallery, because there are already threads for that. We try to find out something here, what ever it is. (I put the important findings on the first page.)