"Quick" And Easy Job Lua Building With AI

Eorzea Time
 
 
 
Langues: JP EN FR DE
Ffxivpro Yellow Box
2712 users online
Forum » Windower » General » "Quick" and easy job lua building with AI
"Quick" and easy job lua building with AI
First Page 2 3
 Bahamut.Surealistic
Online
Serveur: Bahamut
Game: FFXI
User: Sureal
Posts: 16
By Bahamut.Surealistic 2026-06-23 10:03:07
Link | Citer | R
 
I was asked to make a guide of sorts on how I build my Lua sets for all my jobs, so here it is.

I'll start with a TLDR as a brief overview.

I use Google's notebooklm.google.com.
I have linked all the job guides from bgwiki.
I downloaded all my gear with Gearswap using //gs export all, and took that info and copy/pasted it into a notebook as copied text.
I linked the custom gearswap I use (name redacted for SE reasons) on GitHub.
I have a set of rules I have told Notebook to follow.
1. JSE (Job-Specific Equipment) Tier Mapping: All JSE shortcuts must be mapped to their highest possible tier (+4 for Artifact and Relic, +3 for Empyrean) directly at the top of your Lua files, regardless of whether you currently possess those upgraded pieces in your inventory.
2. CSV Inventory Verification: Every non-JSE piece of gear used in your sets must be strictly cross-referenced and verified against your provided Vana'diel Exported Item Manifest. If an item is not in your inventory, it cannot be used.
The Warp Ring Rule: The Warp Ring must be hardcoded into all of your idle sets.
3. Movement Speed Gear Rule: The optimal movement speed item for your specific job (e.g., Fajin Boots for Puppetmaster, Carmine Cuisses +1 for Paladin, Jute Boots +1 for Black Mage) must be applied to your idle and Kiting sets.
4. Ambuscade Cape Rule: It is assumed that you have the proper Job-Specific Ambuscade cape variants (e.g., Rudianos's Mantle for Paladin, Visucius's Mantle for Puppetmaster, Taranus's Cape for Black Mage) with the optimal augments required for the specific set being built (such as Fast Cast, Weapon Skill Damage, or Double Attack).

I then take an example of a job Lua and delete all the gear, just the gear, not the set names. Copy all, or half of it, as Notebook has a limit on how much can be in one text box, and tell Notebook to fill out the sets for (insert your job here).

I take what it gives me, and I paste it into my Lua, and bam, full Lua with your gear for each job.



ok, so now the full walkthrough.

1. Go to NotebookLM and create a notebook for your Character.


2. Go to Job Guides and link all the job guides you want, or all of them as websites that actually have info. If they are outdated, still add them; they may have some random gear piece that no other guide has



3. Link the gearswap and any relevant gearswap setup GitHub

4. Export all of your gear with GearSwap. //gs export all Open the file generated in your Windower/Addons/GearSwap/data/export folder and copy all the information there. Don't worry about random items; Notebook will ignore them. Go back to Notebook and add another source, but this time go to copied text instead of websites
Paste your inventory in there.


You should basically, at this point, have a Notebook with all job guides, your gear, and any relevant gearswap GitHub, and it should look something like this



5. Insert rules:
  • JSE (Job-Specific Equipment) Tier Mapping: All JSE shortcuts must be mapped to their highest possible tier (+4 for Artifact and Relic, +3 for Empyrean) directly at the top of your Lua files, regardless of whether you currently possess those upgraded pieces in your inventory.

  • CSV Inventory Verification: Every non-JSE piece of gear used in your sets must be strictly cross-referenced and verified against your provided Vana'diel Exported Item Manifest. If an item is not in your inventory, it cannot be used.

  • The Warp Ring Rule: The Warp Ring must be hardcoded into all of your idle sets.

  • Movement Speed Gear Rule: The optimal movement speed item for your specific job (e.g., Fajin Boots for Puppetmaster, Carmine Cuisses +1 for Paladin, Jute Boots +1 for Black Mage) must be applied to your idle and Kiting sets.

  • Ambuscade Cape Rule: It is assumed that you have the proper Job-Specific Ambuscade cape variants (e.g., Rudianos's Mantle for Paladin, Visucius's Mantle for Puppetmaster, Taranus's Cape for Black Mage) with the optimal augments required for the specific set being built (such as Fast Cast, Weapon Skill Damage, or Double Attack).


These are the rules I came up with that best fit what I want from this. Experiment and update as you see fit for your needs.

6. The gearswap setup I use has default or example luas for all jobs. If yours has that, take the relevant one and delete all the gear, but not the sets out of it.

Copy as much of it as will work in one text box in Notebook and tell it to "fill out these sets for (insert job here), (insert copied lua)"


The first thing you get back should be something like this


Copy everything from gear.Artifact ={} to gear.Empyrean.feet = "Fili Cothurnes +3" and paste it at the top of the Lua you are making.


I have always set all reforged gear to max, then adjust down to what I actually have. This also always uses reforged gear regardless if i actually have it, i do this so its less changing later on when i do get gear i dont have now. Obviously, you can change this in the rules however you want, but having all the reforged gear at the top makes it super easy to change when you upgrade a piece, just change the +1 to a +2, and every use of it is changed throughout the whole lua.

Then copy and paste the rest of the Lua over the gearsets it just made






This is my first guide; any input would be appreciated.
[+]
 Lakshmi.Avereith
Offline
Serveur: Lakshmi
Game: FFXI
User: Lilianna
Posts: 1318
By Lakshmi.Avereith 2026-06-23 10:20:56
Link | Citer | R
 
But but

Dunna
[+]
 Bahamut.Surealistic
Online
Serveur: Bahamut
Game: FFXI
User: Sureal
Posts: 16
By Bahamut.Surealistic 2026-06-23 10:21:49
Link | Citer | R
 
I do not have brd geared at all lol, I just grabbed a random job
 Lakshmi.Avereith
Offline
Serveur: Lakshmi
Game: FFXI
User: Lilianna
Posts: 1318
By Lakshmi.Avereith 2026-06-23 10:40:16
Link | Citer | R
 
Offline
Posts: 303
By Genoxd 2026-06-23 11:06:34
Link | Citer | R
 
This is why you cannot ask AI to do this with weak rules like this. It makes ***up and will cut corners all day.
You're just not going to get good results unless you bound it to checks. You'll need to ask it to create tests for sets to validate if the set has real items, are the items in the right slot, can they actually be equipped, etc.
[+]
 Asura.Suteru
Guildwork Premium
Offline
Serveur: Asura
Game: FFXI
User: Suteru
Posts: 528
By Asura.Suteru 2026-06-23 11:24:18
Link | Citer | R
 
Where's my minus button for posts
[+]
 Bahamut.Surealistic
Online
Serveur: Bahamut
Game: FFXI
User: Sureal
Posts: 16
By Bahamut.Surealistic 2026-06-23 11:26:37
Link | Citer | R
 
sorry, it made a mistake, guess what, humans do too

it works really good, but sorry for trying to help people
 Carbuncle.Nynja
Offline
Serveur: Carbuncle
Game: FFXI
User: NynJa
Posts: 7632
By Carbuncle.Nynja 2026-06-23 11:45:34
Link | Citer | R
 
"it made a mistake" is something that comes up after a couple people have used it and bugs are found, not right in the preview pics.
 Asura.Psylo
Offline
Serveur: Asura
Game: FFXI
User: psylo
Posts: 461
By Asura.Psylo 2026-06-23 11:46:02
Link | Citer | R
 
Its not a problem to help people, but its works better when people learn to do thing, and sorry but made a lua, understand set etc is one of those things.
 Shiva.Thorny
Offline
Serveur: Shiva
Game: FFXI
User: Rairin
Posts: 3956
By Shiva.Thorny 2026-06-23 11:46:33
Link | Citer | R
 
Bahamut.Surealistic said: »
it made a mistake, guess what, humans do too

Humans learn from mistakes. If you're of reasonable intelligence and spend a few hours learning how both the stats and luas work, you might mix up gear from time to time but you're probably not going to put Dunna in your BRD's sets. Since it's the song midcast effect, that swap alone is probably putting your performance below someone who just fulltimes their buff set because you're losing song+, wind skill, and access to honor march. No moonbow whistle either.

The entire reasoning behind using GS instead of a simpler system is that you can squeeze out every drop of performance. If you offload your choices to a system that doesn't understand that performance, you're probably not gaining anything. Just make some equipsets and call it a day: despite what people say, it's a perfectly viable way to do all content besides new MTs.
[+]
Offline
Posts: 5761
By RadialArcana 2026-06-23 11:51:25
Link | Citer | R
 
Gearswap is needlessly complicated for new players, there does need to be some system to auto build sets from a program or whatever. It also needs to put sensible limits on them too, people make outlandish sets that hammer the server with all kinds of needless swaps. Lots of people using gearswap in an instance is effectively a minor ddos when it's all built on 56k limits.
 Shiva.Thorny
Offline
Serveur: Shiva
Game: FFXI
User: Rairin
Posts: 3956
By Shiva.Thorny 2026-06-23 11:54:49
Link | Citer | R
 
I think gearswap is complicated because the equipment swapping system and the nuance of ffxi's damage calculations are complicated. Players who don't want to learn those things will probably not play at the highest level. If you're of below-average intelligence your best chance of competing is to get a premade lua and go farm or buy all the gear already in it. And, that's what a lot of people seem to do.

I don't really see a great solution. People get into the game for the feel of it but don't have the time, intelligence, or desire to learn all the stat interactions. Players using these tools have inflated the difficulty of content to be scaled for their use. The desire to have something automatically set up the best sets based on what you have makes perfect sense in that context, but this isn't it.
[+]
 Asura.Eiryl
Offline
Serveur: Asura
Game: FFXI
User: Eiryl
By Asura.Eiryl 2026-06-23 11:55:30
Link | Citer | R
 
Gearswap is cake. It's a mountain of absolute *** you add to it that makes it complicated.

Nothing easier than taking a montenten gearswap and filling in gear. Blindfold easy.

But it doesnt autoengauge and autofollow and autocast and autows and autofood and automed I have to actually do stuff QQ
[+]
Offline
Posts: 5761
By RadialArcana 2026-06-23 12:00:45
Link | Citer | R
 
Quote:
Players using these tools have inflated the difficulty of content to be scaled for their use.

Actually I take back what I said, the less people using gearswap the better for the game. I don't blame whoever made it / spellcast cause they had no idea how crazy it would become, but it's absolutely turned into a terrible thing from the pov of view of content creation and server stability.

Pandoras box of ***.
[+]
 Leviathan.Kozumi
Online
Serveur: Leviathan
Game: FFXI
User: Shiyo
Posts: 45
By Leviathan.Kozumi 2026-06-23 13:53:29
Link | Citer | R
 
Bahamut.Surealistic said: »
sorry, it made a mistake, guess what, humans do too

it works really good, but sorry for trying to help people
More AI slop that doesn't even work properly is the last thing the internet needs.
[+]
Offline
Posts: 245
By zeta 2026-06-23 13:54:25
Link | Citer | R
 
RadialArcana said: »
Quote:
Players using these tools have inflated the difficulty of content to be scaled for their use.

Actually I take back what I said, the less people using gearswap the better for the game. I don't blame whoever made it / spellcast cause they had no idea how crazy it would become, but it's absolutely turned into a terrible thing from the pov of view of content creation and server stability.

Pandoras box of ***.

I dont think SE evers build content around players that use gearswap. Gearswap can be as complicated as you want it to be. Not hard to write a simple job file with simple logic.

However I absolutely love gearswap and happy it exists. Personally love to tinker with sets and like adding in stuff like item usage reports that lets me know how many of an item i have after use. Or mode changes.
Also to my understanding GS isnt hard on the server because GS. Isnt lot of it due to the movement swaps some people have? GS itself only sends the final equip() after all rules have ran for the events?
Offline
Posts: 245
By zeta 2026-06-23 14:01:52
Link | Citer | R
 
Leviathan.Kozumi said: »
Bahamut.Surealistic said: »
sorry, it made a mistake, guess what, humans do too

it works really good, but sorry for trying to help people
More AI slop that doesn't even work properly is the last thing the internet needs.

I have a love hate relationship with GPT at time point. At first months ago when I first started messing with some coding tasks it was terrible. However I think now the model and info i fed it over time has improved it.
I am by no means a real coder but I know enough to be dangerous. I also dont have a very good imagination of what I want and sometimes as I prompt it ends up giving me ideas I never would have considered or how best to bring to life.

Not saying AI is good or bad but when it does work right it is fun. (sorry if i am the problem)
Offline
Posts: 4044
By Blazed1979 2026-06-23 14:21:54
Link | Citer | R
 
+1 from me, dude.

These forums have become more and more bubbled over the years, and the same small group of bitter, socially broken posters seem to dominate every discussion with negativity and toxicity. Anything new, or even slightly challenging to their fragile sense of superiority in a 25-year-old game = the type of childish drivel demonstrated above.

Keep innovating. Keep experimenting with AI. The game needs more people trying things, not fewer.
[+]
 Shiva.Thorny
Offline
Serveur: Shiva
Game: FFXI
User: Rairin
Posts: 3956
By Shiva.Thorny 2026-06-23 14:27:58
Link | Citer | R
 
The issue isn't 'AI' or 'trying new things', it's a completely unbounded model that vomits out garbage because the person who set it up doesn't understand how AI works and the AI doesn't understand how FFXI works.

A functional tool that does the same thing *could* be made using AI, but it would use a much different model. Addon or other tool exports all valid items for job by slot. The model would be equipped with skills to evaluate how well sets do common things (magic damage, ws damage, and tp damage can all be estimated with reasonably light python scripts). The cost of recursively checking every combination is too high, but comparing a few options and seeing that a piece typically wins will be sufficient to mostly hit the best options. These are things that can be done using well-established techniques.

JA pieces with no tradeoffs should be hard-defined. There are only a couple hundred of them, it's not an insane amount of manual data-entry. People have spent much longer on these half baked solutions than it'd take to make universal includes for things like that.

Taking a list of equipment and comparing it against guides loosely isn't an approach that will drive optimal outcomes. There's no meaningful way for the model to interpret how pieces interact and verbal directives won't ensure it properly handles key things like haste cap (or song+ on midcast...).
[+]
Offline
Posts: 4044
By Blazed1979 2026-06-23 14:31:26
Link | Citer | R
 
Bahamut.Surealistic said: »
This is my first guide; any input would be appreciated.
use co-work on claude or genspark.
for Claude, set the outcome you require. It WILL achieve the outcome you want.

example - I'm tired of typing in all the POL/SE IDs and PW's for all my 12 accounts. Asked co-work to come up with most practical and time saving solution. Maybe someone's done it, i don't know. doesn't matter. One click button on my desktop now, launches all 12 characters across 3 screens. That was phase 1.
Phase 2: Asked co-work to create a means to create 3x pts of 4 characters - pt composition based on character names.
In game command from my main character "Launch all" - boom. all other 11 characters logged in. 5 mins later parties formed.
up, running as intended.
Phase 3: main character issues commands in LS chat during battle, relevant actions taken to interupt their automated conditional routines.
Co-work is currently learning how to move characters through quests.
I've never done any coding outside of basic gearswap.
Offline
Posts: 4044
By Blazed1979 2026-06-23 14:33:29
Link | Citer | R
 
Shiva.Thorny said: »
AI doesn't understand how FFXI works.
which A.I and when? what part of FFXI? what kind of A.I?


We've automated consulting and built A.I board advisors, finite limited set of mechanics that is FFXI ain't got ***on that.
 Shiva.Thorny
Offline
Serveur: Shiva
Game: FFXI
User: Rairin
Posts: 3956
By Shiva.Thorny 2026-06-23 14:39:15
Link | Citer | R
 
You're using generalities and bragging about clunky tools that existed 15 years ago. 'learning how to move through quests' is abstract enough to mask that you've clearly not gotten the results you're hoping for from it.

The issue isn't that AI cannot comprehend how FFXI works. It absolutely can, but no existing model has all of that already built into training data. That won't be accomplished by throwing a bunch of outdated user guides into a low end model. High context window models can certainly work within boundaries like haste cap, but they don't have a dataset of reliable inputs to get to what the user wants.

Like I said, it's possible to make this kind of tool with AI, it just takes a different approach. Give the AI well-defined skills that allow it to put a numeric value on a complete set, and it can absolutely test out your gear and find one of the best options for that set from what you have. That still requires a more direct hand in the creation of those skills.
[+]
 Shiva.Thorny
Offline
Serveur: Shiva
Game: FFXI
User: Rairin
Posts: 3956
By Shiva.Thorny 2026-06-23 14:46:56
Link | Citer | R
 
Blazed1979 said: »
We've automated consulting and built A.I board advisors, finite limited set of mechanics that is FFXI ain't got ***on that.

It's an interesting comparison, because obviously an AI board advisor operates on much higher stakes than a gearset decision for FFXI. I don't doubt that you've been able to offload a ton of work to your AI tools and your consulting has become exponentially more profitable. However, it's also much harder to objectively say whether their advice led to an optimal outcome. There are too many moving parts to fully evaluate whether it gave the best decision, but as long as it followed industry accepted practices it's probably going to be above criticism.

Not really the same here, we can look at the output and decide if it was wrong based on finite criteria.
Offline
Posts: 4044
By Blazed1979 2026-06-23 14:47:04
Link | Citer | R
 
Shiva.Thorny said: »
You're using generalities and bragging about clunky tools that existed 15 years ago. 'learning how to move through quests' is abstract enough to mask that you've clearly not gotten the results you're hoping for from it.
probably. But i'm not a coder and entirely alergic to it. If a few mins or hours of playing with co-work gets me the results I needed, without the required expertise, that's the entire point.
Shiva.Thorny said: »
'learning how to move through quests' is abstract enough to mask that you've clearly not gotten the results you're hoping for from it.
Not yet. But it's an iterative process.
Shiva.Thorny said: »
but no existing model has all of that already built into training data
again not yet. but once someone does, it is easy to replicate. literally just ask it to provide you with a thorough prompt.
Shiva.Thorny said: »
but they don't have a dataset of reliable inputs to get to what the user wants.
what about the damage simulators? and what does the user want in this case, automated lua generation? at first, sure.
Shiva.Thorny said: »
That still requires a more direct hand in the creation of those skills.
 Shiva.Thorny
Offline
Serveur: Shiva
Game: FFXI
User: Rairin
Posts: 3956
By Shiva.Thorny 2026-06-23 14:54:02
Link | Citer | R
 
Blazed1979 said: »
what about the damage simulators?

Packaging damage simulators into skills is exactly what I was referring to by taking a more direct hand. There are only so many different core sets (TP, WS, nuke, magic accuracy[bounded by magic skill]). Using a skill for each to deterministically decide which set will perform best will create an environment where the AI can actually test options and verify results.
[+]
Offline
Posts: 4044
By Blazed1979 2026-06-23 14:56:18
Link | Citer | R
 
ah I see. that's actually pretty cool, didn't think of that.
Offline
Posts: 4044
By Blazed1979 2026-06-23 14:57:33
Link | Citer | R
 
Shiva.Thorny said: »
Blazed1979 said: »
what about the damage simulators?

Packaging damage simulators into skills is exactly what I was referring to by taking a more direct hand. There are only so many different core sets (TP, WS, nuke, magic accuracy[bounded by magic skill]). Using a skill for each to deterministically decide which set will perform best will create an environment where the AI can actually test options and verify results.
another way of saying this for people who don't know what you mean by "skills" in this co-work context
"Don’t let the AI just give opinions about which gear set is better. Give it actual damage simulator tools. (turn them into individual "skills" within co-work) Since there are only a few main gear-set categories — TP set, weapon skill set, nuking set, magic accuracy set — the AI can test each option properly and prove which one performs best"
 Shiva.Thorny
Offline
Serveur: Shiva
Game: FFXI
User: Rairin
Posts: 3956
By Shiva.Thorny 2026-06-23 15:02:40
Link | Citer | R
 
From there it's possible to group enemies and buffs in a way that a series of sets can be created and each set can be tested against the sims using those parameters. It's definitely a solvable problem, even at extremely high performance levels (auto switching sets based on target and buffs). If the file is being generated automatically, it could even do cool things that would be too ambitious for manually curated files like establish what accuracy level a swap becomes more worthwhile and rank all your potential sets on a direct gradient based on target rather than 'MAX high mid low'.

My complaints earlier in the thread are specifically about the approach used in this thread, not all AI. But, I think that being able to recognize the correct AI tools, visualize what models can and can't work, and get it all together is still a human skill that you can't entirely replace with prompting. Knowing what dense data sources exist, how trustworthy they are, what approaches might work, has considerable value even if it isn't exactly 'coding'.
[+]
Offline
Posts: 4044
By Blazed1979 2026-06-23 15:05:46
Link | Citer | R
 
gave co-work (low/easy effort - fast respond now) access to this specific page and asked it to provide a prompt for people to test on a basic llm like gpt or gemini.
here's its response (i havent tested it)
[+]
Offline
By Dodik 2026-06-23 15:13:54
Link | Citer | R
 
Quickly go blaze one and stop making these posts.
First Page 2 3