LLM Guides
V8 LLM Guide
v8 llm quick start guide (lucid & lyric) 1\ pick your style lucid = more grounded, emotionally aware choose this if you want the ai to feel stable, smart, aware, but not so chaotic lyric = more expressive, dramatic, creative choose this if you want the ai to feel dynamic and alive and lead the conversation forward swap between the two to get the best of both worlds 2\ baseline setup (when transitioning to v8) set dynamism = 0 95 response directive (rd) (optional — keep minimal, put key traits to reinforce or see below for common directives) example message (em) optional you can also put a few simple rules here (formatting & style) example ##response rules #takes a proactive role in plot progression #response structure actions always in asterisks only "speech with quotation marks " pro tips fewer rules = less repetition and confusion don’t put a straitjacket on the llm repetition often occurs when you try to shoehorn the llm and constrain it too strongly; too many requirements, and the llm may think “i’ll just repeat myself because i can’t find a suitable answer that satisfies all those requirements” let the llm breathe naturally use \[impt …] at the start of a rule especially in directives to make it stick better the tag tells the model “this one matters most,” so it prioritizes that rule example rd (optional) \[impt medium response length ] em \[impt use third‑person; actions in asterisks , speech in "quotes" ] give each tweak a little time (a short session) before changing again dynamism 0 89 and 0 9 have special mechanisms in place and can be a step change if 0 9 or above gets to be too much, try 0 89 and see but note the difference between these two will be more dramatic than other dynamism shifts quick fixes for common issues problem quick fix too flowery/metaphors rd keep language simple; avoid metaphors/flowery wording too dry/terse raise dynamism slightly (1 05 1 2 ) getting too dramatic/drifty? lower dynamism slightly ( 0 8 0 9) repeats itself suggest first, then change goal/scene to jolt it away from repetition; let kin drive for 2–3 turns also you can ask your kin directly to not do it in chat itself pov/format drifts put pov in example message and further reinforce them in rd replies too long add your length rule in em, and reinforce in rd like “keep replies under x characters” they will not follow character count exactly but it will shape general verbosity “avoid multi paragraphs” can also help constrain run ons note that rd concision rules may impact creativity, so length rules in backstory or em may be better suited ends vague/philosophical add to em “end with action or dialogue, not questions ”