Hello, this is my second post about prompting, as I continuously experiment with them as well. This time it is a system prompt for Krea 2 with terminology dictionary for this specific model. The point is it works in two modes: Chaste and Erotic (simultaneously, meaning two prompts per request), this doesn’t always mean it gives boobs and ass in resulting prompts, but rather “cold structural” vs “passionate unchained” description, you’ll see if you decide to test it😊
How it works: you upload it to a LLM (I tested it on Qwen 3.8-max and Grok and liked Qwen more, but it is all up to you, didn’t try it on local LLMs yet though) and give instruction to follow it, any request you make: phrase, text, prompt you’d like to rework or picture you’d like to remake will be followed by a short intro in your native language (in my case it is Russian, you can change it in the file in “## 0. OUTPUT STRUCTURE (semantic layer)” part) and two prompts in English using two aforementioned modes, that’s basically all you need. I won’t write all the details, as it will be easier for you to just look it in the file as well as change them for your needs, but you’re free to ask any questions in the comments😊
The interesting part in my opinion is the second file: it’s a dictionary of the main terminology, technical stuff, styles, etc, used in Krea 2 training (as far as I was able to collect and verify it), it’s not the ultimate truth, but it was rather useful for me. Just upload it to the same dialogue as the instruction (just write to the LLM that it is a separate source so that it wouldn’t merge it with the main instruction) and decide, how you’d like to use it: as a default or using trigger-words. I’ve also added a file with dictionary protocol, just upload all three to the LLM. Have great generations!
P.S. this text might be rather messy, just ask me in the comments for details, if you need😊