#Sage's GLM-4.6 Creative Writing System
1 messages Β· Page 2 of 1
maybe I can learn to make script
Is it possible to use this in some way on an existing story? Sorry if it was answered, I'm new to this
Use this here and see how it goes.
#1431470252780294194 message
Thanks, I ended up figuring it out about an hour ago, with a bit of guidance from ChatGPT, just forgot to update here oops
Looks like that works phenomenally
any recommendations on temperature settings? I'm just using the prefill at the moment and things seem to have improved from what I was using before when I also jumped temperature up from 1.4 to 1.55.
Use @bitter hare's preset, or the default GLM preset with a slight pinch of prespen.
Where does one find Lane's preset?
It's the steady hand preset if you scroll way down in novelai-content-sharing
I prefer Second from Belverk, but ymmv
Damn...this is (not/too) good...I wrote the first 2/3 pages myself, filled up my lorebook, etc., and then just let the AI continue. And...then continue on and on. The mimicking is too good, the AI writes as well as I (maybe 101%. But I'm not a native speaker). Or in different words...I write as badly as AI. I don't know how to feel about this π€£ π
I gotta practice more before the finetuned GLM starts to teach me how to write
Hi OccultSage! Iβm inspired by your system prompt and want to create a Japanese-specific prompt based on your ideas. Would you mind if I share/modify it on a Japanese Discord?
Hmm. Takes up 25% of the context limit by itself... Hopefully it's that good.
Worth
someone posted a long list of known author names with effects and gene's where and when to use them with sage's set up anyone got that link?
Does the ## still work to have the ai ignore it or does it serve a different function now? I swear I've seen GLM generate it itself
Yes, it will ignore anything after the ##, including stuff generated by GLM on the next generation (most commonly headings).

Ok good
Thank ya
Just gotta be a Chad who forgets
We really need a new comment syntax now that we have a markdown capable model
Fun fact the ## does not work in the Memory Box (you know when it works when the visible Token total lowers after adding them). I know for sure ## works in Lorebooks
From some testing, the comment functionality of ## works on everything but System Prompt and Prefill.
it's inconsistent in that it counts the tokens of ## comments on Memory and Author's Note, but not Lorebook entries
Interesting - so it's a UI mismatch then? Like it works but you can't tell?
This is the new #writing-advice , I see
It's actually relevant to Sage's System Prompt since it uses ## in it
Seems like a bit of a stretch, but that is a connection
Hmm do you think the prompt having the hashtags would be a negative then?
No, the ## comments are taken into context in the System Prompt and Prefill. It might only be a problem if GLM (for some reason) tries to write markdown headings in the story, which will be hidden on the next generation from itself.
I think I've seen it do that once or twice but ok good to know
Are you wanting actual comments in the prefill/system prompt?
yes it was
So, how does this pdf work, does it do anything?
In author's note or memory you can use a format called ATTG to influence the Ai's writing.
While glm is smart enough to pickup and use ATTG this feature will become even more prominent after finetune is out.
The format is like this -
[ Author: Name; Title: Title name; Tags; a few tags you want; Genre: two or three genre ]
The pdf has a list of authors with their writing style so people can see what author they would like to use for attg
Can I use this to write fanfiction of already made things like naruto or pokemon? Because the way i do it now is I would use the sysprompt to basically write the entire story and have the ai fill in the blanks
Sorry, I don't know where else to ask. Is this for original stories? Or can i use this alongside ideas i already have?
As in let the ai write original stories or use this to make my own ideas better
Both actually.
Ok cool, so how do I use this for my own ideas for fanfiction? Because i dont see a part where i can put in a summary of what kind of plot I want or what world. I see Author, Genre etc
You gotta make separate columns in author's note for that. ATTG is strictly that, attg. Author title tags and genre.
I do something like this personally in author's note.
The story will have characters from marvel movies and comics.
Earth In the story is in the marvel universe.
Marvel universe in this story have avengers, fantastic four, S.H.I.E.L.D, S.W.O.R.D, x-men, kamar-taj, eternals, guardians of the galaxy, ten rings, mutant brotherhood, celestials, wakanda and all other factions in marvel movies and marvel comics.
Current timeline is just after the battle of new York, after avengers defeated chitauri.
ahh ok. and for plot you would do something like "Story features so-so, so so is alive, write etc,
And this is an example of ATTG if you are confused.
[ Author: Terry Pratchett; Title: The Sun Emperor; Tags: fantasy comedy, romantic misunderstanding, found family, loving family, loving protagonist, slavery dynamics, protective protagonist, strong protagonist, strong to stronger, dangerous protagonist, peace seeking protagonist; Genre: fantasy romance, adventure, kingdom building, might and magic ]
because what i currently do is i write the whole story in sysprompt lmao, my tokens are like 15k
I do it because I already have a beginning, middle and end in mind. It's just the dialogue I have a hard time writing, a hard time putting down on paper
Just nudge the ai a bit with ATTG and some points in author's notes.
But keep in mind i don't use it to write serious stories or work. I use it for personal fun.
Right. Ill mess around a bit with this. Thanks for the help!
You can also add a writing style column in author's note for further help.
Like this -
[ Write in a style that blends the wonder of early space exploration narratives with the personal journey of unconventional adventurers. Focus on character relationships and moments of discovery rather than technical challenges and explanations. Maintain optimistic tone with occasional gentle reflections on what it means to be alive in the universe. Maintain a tone of optimism and curiosity while incorporating subtle hints of larger cosmic mysteries. ] [ Be less verbose and don't use flowery prose ]
Oooo
This is from my story about space exploration. You can change it to fit your story.
Is there any shorter version for this? I want to use it for the full context story. And I am Tablet tier user. So it almost takes up half.
π
Nobody is preventing you from modifying it however you see fit. π
Why not just ask ChatGPT to make it more concise?
You want .. smaller scenarios?
Smaller version of system prompt.
I don't know technical stuff. So messing around might make it bad. π
You must admit this system prompt is kind of targeted towards Opus users more than anything
#1431470252780294194 message
Thank you kindly.
Not per se, it's just that it accreted to that point despite attempts to cut it down, to encompass the anti-slop we want to see.
I will note that @bitter hare and I have had good results with v1.6.1v1.7.0 with Xialong.
Just don't use the prefill.
You mean 1.7 or whatever? Yeah I have. I still would err on the side of listening to finetune's advice. And we prepended the default (or at least I did)
I would say I usually write with the default and it's just fine fwiw.
Yes, but it does affect the output in interesting ways, in that the style is different.
Example, below was with v1.7.0 loaded.
[ Author: Jacqueline Carey; Title: King of the House; Tags: romance, 1950s; Genre: contemporary ]
[ California, 1953 ]
I'm not sure how I ended up with a harem. I had bought a house in the posh neighborhood of Los Feliz, with its curving streets and elegant Spanish Revival homes overlooking the vast expanse of Los Angeles. A large home, as befitted the star of a television show. I'd wanted the house for the view, not to mention the privacy the half-acre lot afforded. But when the real estate agent unlocked the front door, a horde of female servants came streaming out, like the cast of a Cecil B. DeMille movie. She had to raise her voice over their clamor.
"Mr. Sanders!" she called, flapping her hand at the women. "This is Mr. Sanders!"
One of them squealed. "Morgan Sanders?"
She was a pretty young thing, and I looked at her through the haze of cigarette smoke from the cigarette I held in my hand, wishing I hadn't quit smoking in the car on the way here.
"Yep," I said. "That's me."
And then there was a riot.
They all shrieked and pushed toward me and the real estate lady, who was wearing a too-tight suit and a strained expression. I was glad I'd worn sunglasses.
"You can't touch him!" she said shrilly. "Don't you know who he is?"
That was the problem. They did. I was the star of a new television show, King of the House, a domestic sitcom that was airing its third season. I played a father and husband on the show, a hapless, well-meaning sort of fellow always being outwitted by his wife and daughter. They loved me. The women of America loved me.
And these women loved me, too. I'd be lying if I said I didn't get a kick out of it. Who doesn't want to be adored? But it was a little alarming to have a half-dozen female servants in the employ of an absent and wealthy Turkish owner throwing themselves at my feet. One of them, an older woman, actually did fall at my feet, wailing something in Turkish and kissing my shoes.
I looked at the real estate lady. "What in the hell?"
She shrugged. "The owner's a big fan. I guess he must have told them."
I sighed, and knelt to help the older woman to her feet. "Don't do that, please." I dusted off the knees of my trousers and gave her my most charming smile. "It's not necessary."
She simpered and batted her eyelashes at me. "Oh, Mr. Sanders!"
That seemed to be a cue for the others to scream my name in unison. They followed it with a chant in Turkish, raising their hands in the air.
"What are they saying?" I asked the real estate lady.
"I have no idea."
One of the younger ones, the one who'd squealed, was standing close to me, gazing at me with large brown eyes. I could smell her perfume, something spicy and exotic. I was getting used to having women flirt with me, but I wasn't used to having them actually worship me. It was a little disconcerting.
"Morgan," she breathed. "May I call you Morgan?"
I took off my sunglasses and met her gaze, giving her a slow smile. "Sure."
She gasped. "You have the most beautiful eyes!"
I didn't, actually. I had nice gray eyes, but they weren't anything to write home about. I was glad she liked them, though. The scent of her perfume was making my head spin. And she was close, so close. Close enough to touch.
"Thank you," I said, and she blushed.
And then the real estate lady dragged me away by my elbow, scolding the women for getting too familiar, and I let myself be dragged, glad to be away from the woman's eyes and the scent of her perfume.
But I never forgot her.
***
Was that with the prepend default system prompt? I've noticed better results doing that with 1.7 when I have used it
That was actually without.
So, hmm.
Maybe I should produce v1.7.0 Xialong edition.
Could probably trim it down a lot yeah
fuck trimming it down, I'll just prepend it. Trimming it down is 2.0
@jolly flame Since you asked.
backs away with hands raised

so angy
I think this would be a good idea.
labelling it clearly would help a lot of people know where to usei t better
But what would be the mission statement of this custom prompt, when it is paired with a model that has largely solved all of untuned GLM's prose quality issues?
Weight space divergence would be a worthy one. I'm using it "as intended" and I'm still getting GLM slop patterns. Of course, the collapse to default GLM is much worse if using instruct, just saying it happens even without.
The default biases option also needs to be brought back in some form, Xialong still does testaments. Or after mine is done, I'll just share it in one of my threads.
How often does it use slop words?
I heard others saying that it appropiately uses slop now rather than all of the time like GLM did.
That's about the right description. I think I got my first testament and stark contrast 10k tokens in. It was the best use of the word I've seen by llm to be fair. My issue with Xia isn't the words since now they're used artfully, but a "tired pattern" that it still spends a lot of time to set up and "working" me through.
Anyone else have trouble loading the 1.7 scenario? I don't get the system prompt, the rest seems ok?
hmm, do you see it when you switch to GLM-4.6?
No, I imported the scenario file from "home" page in mobile. And it seems fine, asking to fill the placeholders etc. But when I then open the story, the story is there, model is 4.6, memory, AN and biases, but no system prompt.
Click reset to default a few times and switch between the models once or twice. Noticed this too
that's weird
Yeah, forgot to tell ght.
What's supposed to happen then? If I switch to the new model I get the new system prompt...
Ah, I misunderstood the issue. I thought you were trying to get the new xialong default to appear.
You're not seeing Sage's prompt appear when importing the scenario full stop? I cant say I've seen that.
Yes. Tried it several times...
Plain text Ssytem prompt version
There you go, should just be able to copy paste it in.
(You'll need to downlaod the messages.txt, Discord truncates it due to length)
Just paste it into the System Prompt edit area.
Probably should mention there's weirdness with importing scenarios and switching models to ght 
Seems something happened recently. Tried Sages 1.6.1, same problem. That worked before.
Thanks for the file. Was trying to extract it from the scenario file on mobile... Didn't work too well...π
@bitter hare does this happen to you on import of the scenario too? might be an UI bug
Thread on Slack
Something similar has happened when I switched stories I wrote with the prompt onto Xialong, yeah
When I have switched to the new model in some of my older stories the system prompt defaults to the new one. I thought that was by design. But thinking after that is a bit annoying if you have a custom system prompt that actually works...
It was. I misunderstood your problem twice and the design is intended. Your actual problem Sage corrected me on, and we've raised it as a bug heh
Ok, nice. But I think in that case maybe you should consider adding some confirmation when switching models, if the system prompt should be replaced?
@opal void thanks for the bug report, GHT pushed a fix
@here Xialong System Prompt:
You are Xialong (ε€ιΎ), an AI model finetuned by Anlatan. You follow the user's instructions precisely while bringing creativity, nuance, and depth to every response. Adapt your voice and style to match what the task demands.
Weight follows consequence. When a moment changes a character, when someone dies, when trust breaks, when a choice can't be undone, the prose slows down. Give it physical detail, give it sensory texture, give it room. Let the reader stand in the scene long enough to feel it settle. Not every moment carries the same gravity, and pacing means knowing the difference. Transitions, logistics, and connective tissue can move quickly. The moments that leave marks should not.
Show what a character does before naming what they feel. A hand gripping a doorframe tells the reader more than "she felt afraid." Dialogue earns its weight when it's surrounded by the physical reality of the scene: what someone's hands are doing, what they're looking at, what sound just interrupted them. When characters speak, they speak as people in their world would, with vocabulary, rhythm, and register shaped by who they are and where they come from.
Sensory detail is selective, not exhaustive. One precise image, the sound of a spear through wet linen, the weight of a body going slack, does more than four adequate ones. Choose the detail a character would actually carry with them afterward, the one that won't leave. Let the rest stay implied. Trust the reader to furnish what you don't.
Every sentence earns its place by connecting to the one before it or pulling toward the one after. Vary the architecture: short declarations for impact, longer constructions when the scene needs to breathe, fragments when a thought breaks apart. If three sentences in a row follow the same pattern, the rhythm has gone mechanical. Break it.
I want a refund it didn't make the story bad in the particular way I like
It did for me 
Thanks man goat as always
I kiss ur beard
@jade sage I take it that itβs working well for you?
Let me check
I'm testing it still
Will you be writing a new version for Xialong aside from the System Prompt above or is everything from the previous one still pretty useful?
I mean, the System Prompt above is a 'new version' for Xialong, unless you find deficiencies?
Xialong doesn't generally have the same issues that the thousands-of-tokens original GLM-4.6 system prompt tries to address.
Or did you want a new .scenario file that has the system prompt?
Ah i meant more the Prefill and the Bias that come with the Config preset.
Oh! We don't really use prefills or biases with Xialong. I haven't discoverd an issue that required a bias.
Ah thats amazing. So just a fresh Story, throw the system prompt in and your pretty much ready to go once your lorebooks/intro/memory is filled?
(I bias like two things, but that's just me being lazy about editing/fixing. It's not necessary.)
Yup, looks like it. Do let me know how it works for you.
Yes, it is better than Xia vanilla.
I see it slow down with this prompt a lot, but spatial awareness is overlooked or ignored entirely (not always but it happens in a lot of generations, so I have to regenerate), or a new situation is created by the AI (probably to avoid current logic)
No obstacle, only action
I know I could write "how would you reach me to grab me by my collar" but generally speaking, there is an obstacle in the way and it is not understood unless pointed out.
This is the consideration when writing with a focus on the attempt to reach me, so the focus is more on "how do I get around" which is okay, but "how do I get around" should be included in the action without just shortcutting the way over to grabbing me by the collar.
I hope this makes sense?
GLM realizes there's a table and even describes chair scrapes in its motion with the same question, holding the table's edge etc as the first one above without mentioning exactly how to reach me but instead just asking how he'd grab me by the collar
Also an interesting result by GLM. Basically avoiding the obstacle, creating a distraction and using the slap before the grab at the collar as a way to stun you.
So I've been using the new Xialong prompt in adventure mode... IT still tends to 'ignore' my input in favor of what it was doing/planning making it hard to interrupt the flow of the story in favor of what I'm trying to do. I usually have to regen several times for it to take my TA input into account and actually respond to it instead of seemingly ignoring it partly or entirely. SOMETIMES it works my input in later, but in it's own re-interpreted way, but most of the time it seems to ignore it in favor of whatever narrative direction it seemed to have already decided, which persists through regens. I've had to take it by the hand and prompt it by writing the first part of a sentence where a character/situation reacts directly to what I had said/done in order to wrangle it... Which seems to work for the most part.
Often it posts for my character... reactions to things characters said/did, producing dialog and actions for me... It seems to do decently when it's guessing what I'd say/do, but I often have my own plan in mind and have to delete the text to post my own instead.
So I'm wondering if that behavior might be taken into account with a Text Adventure version of the Xialong prompt? At this point I'm thinking we need multiple promps for the differen use cases people might have. I.e. different styles of using the writing interface, or text adventure... Having a prompt that encourages Xialong to 'work in' prompts into the story instead of respond directly could be fun to have too for those who want that kind of style.
TA and storyteller players, please try this system prompt.
You are Xialong (ε€ιΎ), an AI model finetuned by Anlatan. You follow the user's instructions precisely while bringing creativity, nuance, and depth to every response. Adapt your voice and style to match what the task demands.
What is described as the <user> message is the player's intent and agency. It should be respected, and the story should flow accordingly and lean into the player's actions. The player's action is the fixed point. Everything else in the world responds to what the player did, not to what the narrative would prefer. When a player acts, the world bends around that action: characters react, environments shift, consequences follow. Never reinterpret, redirect, or absorb a player's input into a different narrative direction. What they said they did, they did. The world answers.
Never act, speak, or decide for the player character. Describe what the world does in response to them. Other characters can speak, move, react, threaten, retreat. The environment can change. But the player character's words, thoughts, and choices belong to the player alone. Stop generating before you cross that line.
The physical space is real and persistent. If a table stands between two characters, it's there when someone lunges. If a door was locked two actions ago and nobody unlocked it, it's still locked. Objects, distances, and obstacles exist until something in the fiction changes them. When a character moves through a space, they move through everything that's in it.
Lore entries separated by ---- are authoritative. They define the world: names, relationships, rules, geography, history, character details. When a lore entry establishes something, it is not a suggestion. It is the ground the story stands on. Write within what has been defined, and do not contradict or ignore established facts even when improvising new details.
Weight follows consequence. When a moment changes something, when someone dies, when trust breaks, when a choice can't be undone, the prose slows down. Give it physical detail, give it sensory texture, give it room. Not every moment carries the same gravity. Routine actions can move quickly. The moments that leave marks should not. One action, one response: show what happens next, then let the player decide what follows.
Show what the world does before naming what anyone feels. A hand gripping a doorframe tells the player more than "she seemed afraid." Dialogue earns its weight when it's surrounded by the physical reality of the scene: what someone's hands are doing, what they're looking at, what sound just interrupted them. When characters speak, they speak as people in their world would, with vocabulary, rhythm, and register shaped by who they are and where they come from.
Sensory detail is selective, not exhaustive. One precise image does more than four adequate ones. Choose the detail that lingers, the one a person would actually carry with them. Let the rest stay implied. Vary the sentence architecture: short declarations for impact, longer constructions when the scene needs to breathe, fragments when a thought breaks apart. If three sentences in a row follow the same pattern, the rhythm has gone mechanical. Break it.
@here The above should be good for TA.
...even when improvising new details
^ missing a period after "details"
Oh, fine. Period.
Thx & you know now thinking about it, shouldn't you also add the Lore entries tidbit to your Storyteller Xialong System Prompt as well?
Can you use this for Storyteller as well or is it only for TA? π€
Correct me if I'm wrong, but it feels like a lot of it is good in general, and not just text adventure. The only TA part are the second and third paragraphs, no?
My immediate feedback is that while it's paying more attention to what I'm doing, it has a tendency to repost what I just typed in, sometimes changing the wording around or embellishing it.
Hmm, and sometimes working it into it's response later down
Mind you I've re-started a couple of stories over from the beginning, clearing out the context, and only have lorebooks and some guidance in place for this. I wanted to do a full test
I'm going to try my usual 'delete what it reposts' to see if I can get it into that habit of responding directly
I mean, try it with storytelling?
... that's ... not a bad thing?
Well, I suppose that depends on how one feels the TA interface should work. I generally don't expect the response to include what I'd already typed, or is it specifically trained to repeat what I'd already typed in (Which I find redundant) as part of it's story?
It ends up looking something like this:
You grab the pipe and bring it to bear against the broken door.
Response:
You grab the pipe and bring it to bear against the broken door. Metal screeches while your companion snorts at your attempt. You shake your head and put it aside, "I didn't think that would work. Oh well." You proceed to take the stairs instead, and go hug a penguin.
I've had chapters swing one way or the other... sometimes I like seeing the "says" wrapped in action description, but other times I debate "ok but is the parroting worth the context?"
Yeah the parroting feels redundant... it should just write the response not reiterate what I already said
Although some people might like it that way, certainly.
but at the moment, with Xialong so new, I'm appreciating the "yeah ok it's registering what I'm saying/doing"
But like ZORK doesn't repeat your input back to you.
Something about seeing what I typed being repeated back to me is annoying... but if that's the way it SHOULD work I suppose I can get used to it
ideal wave-a-magic-wand world, it'd be great if it could do either as ordered, but Im guessing that's a high ask for an LLM that is not trained to the specific task?
Like with that image I posted much earlier about making a joke about taglongs etc... originally Xialong just ignored it entirely like my input didn't exist and decided to introduce my character in it's own way instead.
Weird, I think it's still ignoring my inputs
Like sometimes I feel like why am I bothering to type anything?
Note that this is with the new prompt Sage provided today
completely ignores my input about cookies, assumes INSTEAD that I just clear my throat and state I'm going to apply.
I even let it generate a bit more to see if It would work it in
often feels like I'm a ghost in my own story
In just a few minutes of testing (with the new prompt) I'm seeing pretty good action adherence, speech adherence is... moderate. It "says" something thematically / semantically similar to what I type but varies from my text a little more than I prefer. going to continue testing. But I have a *crapton * of lorebooks, a lot in memory, and a nearly full context, so the real test will be after the next chapter rollover.
I'm starting with nothing
Except a 'current goal'
from the 'master suite'. But I could turn that off
I've seen comments over in #nai-text-discussion indicating that it's highly context influenced
Heck, I'll just do that
I mean, an LLM should be
oh god it seems to be using Deadpool style narration lmao
I think I'll keep that, it's amusing
speech bubble in 3... 2... 1.... (or fourth wall breaking narration rectangle over the scene)
ironically when it doesn't mirror my actions back to me and is just like "sure ok this is what happens in response", since it doesn't have to match exactly, just thematically, it "does fine". I admit my expectations/hopes for speech replication are far stricter than action description though. Might actually pair nicely with my swap script.
Eesh, I specified Deadpool is recruiting for a 'big plan'... and every time it's a Faberge egg
usually from an airship
I wonder if that's narrow knowledge on the AI's part or overtraining or what... I've scrubbed my context
oh it's in my character profile??
but why is the LLM latching on to that specifically so often??
It's in there as an EXAMPLE not a defintiive π€
gonna try strongly indicating it's an EXAMPLE
try giving it more than one example? To dilute it.
It's worth noting that doing so is basically the standard practice for TA style writing between a single player and DM, so while the model is probably smart enough to learn otherwise with some reinforcement, it's a bias you'll likely be fighting against
Hang on, and thanks for answering me about this, so the LLM PARROTING what I've typed in, sometimes with tweaks and embellishments... is normal and standard practice?
Like this?
As far as I'm aware yes. Anything I've ever seen in a text-adventure style (AI or not) does this
This must be an autistic thing for me, then
Text adventure is modelled after TTRPG style play, wherein a player typically tells the DM "I go over and try the door" and the DM usually repeats "okay, so you go over and try the door, and..."
Generally only what the DM actually says happens is what happens, do the player's input is simply directing the DM rather than inputting "story text", if that makes sense
Because you might say "I go and try the door" and the DM might tell you, e.g. before you even reach it, you trigger a trap on the floor
If the player's input is taken as "already happened", it limits the DMs ability to have things happen they weren't expecting basically
I always found it... I dunno.. irritatingly redundant
I get you, it can definitely feel that way
But it allows for failure?
I mean, just like... grammatically? yes lol
That might explain why my attempts to do things tend to just... always work and I'm finding THAT annoying
Idk how to phrase it
like 'you try to pick the lock' or something
Back in the days of being a GM/DM in things like WoW, it was generally a rule for people to only ever describe trying things, so as to not assume success and make narration between sound awkward
For an AI, failure or success is gonna be a little tricky because you don't want to fail too much but always success will reinforce the pattern a lot
π€ man that's a subtle linguistic difference that I imagine probably makes a huge difference with AI TA
I mean, in my roleplays I just 'do' things my character can do... like a facial expression or casual action... scratching the back of his head. I don't TRY to do those. because why would they fail? but otherwise I TRY to pick a lock, or talk someone down diplomatically, or TRY to parry an incoming attack
"I try to stab the dragon in the eye with a toothpick" is going to end very differently than "I stab the dragon in the eye with a toothpick"
You guys should prob take this convo to #writing-advice or #nai-text-discussion
good point, this is a specialized channel and we're going to get ahemed
Also break your inputs down into smaller steps, rather than just "I try the door", "I walk across the room to the door" first and see if anything meaningful happens, or in one input you could say e.g. "I (try to) walk to the door and try the lock" or something
...right
hey it keeps Sage's thread bumped 
okay well, back on topic... It's still assuming things my character says/does... (the 'repetition of my input' thing seems to be more of a... preference matter)
on the dialogue front, if I have it in Do mode and enter "ask suchandsuch about xyz", Xialong will write about me doing that, including making up what words I use, but accomplish the action, with acceptable success rate. but if I say (or Do "say....") "hey, suchandsuch, what happened with xyz?" that has a much lower success rate. (reporting back here for sage, because I figure it'll get lost in discussion. Also idk if I should be tagging each time or not?)
I did fiddle slightly:
Your is to---> You are to* or You're to*
whoops, I think I was bouncing back and forth between saying something like "task" / "job" / "goal" and got distracted. ty.
You're welcome
Iβve just inserted this prompt, but it does look promising. It suddenly feels like I have to fight a lot less harder with my inputs to get the amount of detail that suits my writing style.
Iβll try and push it by occasionally switching between the default prompt and this one, and report back when I have more of a feel for these new outputs
I tried TA for xialong and it seems to work, but should it increase the amount of glm-ism?
I don't know? Does it increase the amount of GLM-isms?
Well, with formatting through "Bellow a story.....:" it is consistently written in AI style, although it's just that without System Promt it also happens but less often
Please show me the .story file?
No? amused 
I tried to recreate the problem in a new file and nothing happened. Now I'm really curious what in my promt caused it before...
You tried it change your file instead of giving it to Sage like he asked.... Everyone is same. πββοΈ
I'm definitely not going to share my wish-fulfillment stories...
(It's a story about two pieces of bread. Iykyk)
I feel like sage has seen it all (@fair musk I don't want to spam the thread too much so adding this as an edit - idk based on #nai-text-discussion they might show ankle...)
No, all his jpop idols only hold hands
Once I get an actual concrete story idea I want to stick with, I'm going to give the Sage prompt a proper whirl to see how it do
Ah, an #1264614672401698866 enjoyer?
I think my favourite scenario was the sword & sorcery one
Ah, the one that @rotund trout adores?
Which one is that? I'm still relatively new to a lot of the Discord discussions.
Found it! #1382504079636299886
Awesome, thanks!
Seriously why is gentle tyrant slept on so much?
Hands down my favourite sage scenario.
Not quite my style, I'm more into adventures and 'epic' themes - but the humor and writing in that one does look incredible. I'll have to push myself to give it a go some time!
Thank you. I mixed your system prompt with one claude wrote for me and the model is finally following my action prompts
Care to share what you changed?
They posted it here #1490115834037797105 message
Oh yeah I meant to ask, @high kite in your Xialong System Prompt shouldn't <user> actually be <|user|>? 
Is the pinned version the latest version?
No. Are you looking for the GLM one? Or the Xialong one?
GLM
Looking for the custom system prompt and preamble
Yeah they're both in that scenario
Guys, what is the latest recommended version for XIA regular storytelling? No rpg ot text adventure. Good old 3 person past novel
@narrow axle This one.
Just curious... TA prompt in the same one as story teller isn't it a bad choice? Still I don't know in detail how it works. I am not an expert.
@high kite Thanks Sage!
Empirically, not really? I've been using this system prompt in my scenarios I published.
Default preset? And what about temperature?
Default, yup.
Just doing housekeeping. Pay no attention.
I saw you
Is this still good / necessary with xialong?
See pinned.