(Many months back, CT5Holy wrote the AI Team Building Guide
, This is an updated version that aims to be more accurate and detailed. While written with FFT: Arena in mind, most general information should be applicable to any FFT AI Tournament. This Guide will have two sections, one for basic tips and AI help, and one for more in-depth ones.)
If you're new to FFT: Arena or AI Battling in general, this thread will help you better understand the AI and what goes into making a successful team. For the Rules on making teams in FFT: Arena, Gear, Abilities, Stats, etc, you can check the Master Guide
, which should always be up-to-date. Don't worry about those until after you've read the Basic Team-Building Tips and Basic AI-Behavior Tips sections, though.Basic Team-Building Tips
0. Observe other teams through the many videos of AI battles, here and elsewhere. Even if they're not FFT: Arena videos, they'll give you a good glimpse into how the AI functions at least a basic appreciation of how to build good and not-so-good teams of various battle styles. While the skills, classes, stats, etc. can all
be different between Vanilla, Arena, and the many other FFT patches out there, core game concepts are universal. Knowing your core game concepts is a key step in being good in any AI-Battle situation, and is what the "Basic" sections of this guide focus on.
1. Pick a plan, and stick to it. Make sure every unit fills a role without being dead weight, and make sure you don't put too much importance on any one unit. A team that can support each other well will do better than a team of four "good" standalone units. You don't need any sort of involved concept, nor do you need stick to a basic rubric like Physical-Magical-Status-Healer, but you do need synergy among your team. A team that isn't well-focused and can't support each other well is usually bound to fail. Don't worry about having a full Primary/Secondary/Reaction/Support/Movement loadout on every unit, because this is secondary to ensuring the unit does what you need it to do. There are many ways to give a team synergy, such as Elemental Absorption, Ability Combos (Zombie + Seal Evil or many others), screen-wide skills such as the Lore skillset being used to heal small values and trigger one's own Save or Up Reactions. It doesn't need to be complicated like that, though - a simple team that does damage and keeps itself alive is fine too when done correctly. Some concepts are definitely a lot more involved than others, but once you've chosen one you feel can be successful, stick close and don't stray too far from it. That's the best way to ensure its success.
2. Don't give your units abilities they don't need to fulfill their roles, and don't feel forced to spend all your JP if your units all do what you need them to and still have some left over. This is very much a part of the previous tip, but in reverse. If the AI has skills you don't need it to have, you run the risk of it wasting turns on it at inopportune times and losing easily won games. It's good to cover as many bases and flaws as possible on your team, but don't extend beyond your comfort zone to do so. Cover too much, you begin losing to stuff you should beat because you lose focus. The AI isn't a player like you - more options isn't always good. Similarly, as I'll detail below, more Move and more Speed also isn't always good. Constructing a team for the AI to use is far different from constructing a team for yourself or another player to use. Remember this at all times.
3. You'll want some form of healing and resurrection on most teams you make. Not every unit needs it, but its usually a good idea for 2 units (and almost required that at least 1 unit) carry some form of it, even if it's just a Phoenix Down. Murphy's Law - everything that can go wrong, will go wrong, and the ability to recover and revive is your answer to that. Make sure you keep this revival on a unit that's not likely to die early, either - especially if you only run one or two units capable of revival. Being able to raise things is no good if the things casting your Raise Dead skills are the first things to kick the bucket.
4. Speed is important, as should be obvious. More turns means more Tempo, which wins games. The AI knows this, hence why Haste is one of its top priorities when its either idle or looking to cast a positive Status. However, don't forget your other stats - if you don't have the MP to use your skills, or the PA/MA to do relevant damage, or the HP to live long enough to even see
those additional turns, then getting more turns than the opponent is obviously not worth a lot. This is why it's important to balance how much Speed you want to add with your other stats. A 9 Speed unit that's constantly threatening is usually better than a 12 Speed unit that barely jabs for 70 damage and can easily go ignored. An important note here is the concept of Speed Synching, or making sure your Speed is well-balanced between the individual units on your team. If one unit runs forth before the rest, it'll likely die far too removed to be revived and spell curtains for the remainder of your team. For that reason, a team with four 8 Speed units is usually better than a team with one 12 Speed unit and three 8 Speed ones.
4.5. A note that specifically applies to FFT: Arena - Haste only boosts Speed by 25% and lasts for 32 CT, while Slow reduces Speed by 50% and lasts 24 CT. This means that, unless you're running an uber-Speed setup to begin with (12+ post-ger Speed), Haste will only give you 2 Speed, and past 12 will only give you 3 Speed. Slow, meanwhile, will undercut Speed by at least 4 no matter the target, leaving anything hit with Slow at 4-5 Speed for the most part. Considering Speed changes are far more noticeable going down than up, this makes Slow very, very powerful when used correctly. Haste is still powerful itself, because the difference between 8 and 10 Speed is very noticeable, but it's not a mandate for every team running 8-9 Speed, and units already running appreciable Speed (10+ base) are probably best off foregoing it entirely unless it's just something you can get without going out of your way.
5. Similar to Speed, Move is obviously important, for similar reasons. The less turns you spend idly Moving into range, the more turns you get performing relevant actions and the more Tempo you gain. Again like Speed, though, max Move isn't always good the way it is in single player, since the unit's other stats still need to be sufficiently fleshed out or their extra Move means little. High Move also means your squad can become quickly divided, even if properly synched, so there's the risk of a unit going down and being unable to be revived before its permanent demise. It also runs the same risk as Speed in that a team with one 6 Move unit and three 3 Move units is generally going to be worse than a team of four 3 Move units - the unit with 6 Move is essentially getting two-to-one turns to Move when compared to its allies, allowing it to make contact long before the rest of its team and get killed long before the rest come into range. Move Synching, while not as important as Speed Synching, is still important to a team's success, so at least be sure the Move differences between units on the same team isn't too great.
6. If you're having trouble figuring out how strong or weak something is, WinCalc.exe is your bestest best friend in the whole world ever. I mean it. Outside of Critical Hits, nothing about FFT's damage is randomized outside of skills explicitly made to be so. It's very easy to number-crunch your expected damage or hit rate in literally any and every possible situation in a matter of seconds. Just remember that FFT always rounds a number down when a fraction appears (outside of Short Charge, which rounds up), and that with Protect/Shell/Attack UP/Magic Attack UP/etc, its your primary PA or MA factor that is multiplied against, not the final value. The Arena-custom Supports Overwhelm and Unyielding (along with Fury/Faith) affect the final damage value instead, but they're the exception, not the rule. Numbers are scary at first, yes, but FFT is fairly simple about them. Even if you don't do the math "right", what you end up with should be close enough to give you the right idea.
7. Arguably, this is the most important tip. Remember that sometimes you just need to get lucky to win
. This is important to know, important to remember, and important to accept as something that is perfectly fine
. No team is flawless, and as stated many times before, trying to extend too much and cover too many flaws will just make it cave in on itself. The important thing is being aware of your team's weaknesses, and knowing which you can and cannot cover. When interpreted in its broadest sense, this tip is true of almost every competitive game in existence
. Even the best player needs a little bit of luck. Sometimes you don't get that luck, or your opponent does, and you lose. It sucks. That's when you examine your team, figure out whether there's something you can do to remove the need for luck in that kind of a matchup without weakening it in other areas. If you can't, you can't, and it's a bad matchup. You do what you can and just hope to be the one who gets lucky next time. This goes right back to focusing on your strategy.Basic AI-Behavior Tips
1. The AI is quirky, and has a tiny bit of randomness to it. It'll usually use the best skill in its arsenal, and is very smart about knowing how to use many skills you'd expect it to be dumb with, such as MP Destruction. The AI isn't "stupid" so much in that it merely has a different set of priorities than a player does. Like I said though, it is quirky and has a very mild bit of randomness. If you're using a status-based unit, such as with Talk Skill, and it decides it wants to do damage on a specific turn, and all it has is a 4 WP Knife for damage options, it'll run up and stab a bitch instead of using Mimic Daravon. Even if the unit isn't geared for damage, remember the AI will sometimes do things like this when deciding what to do to an opponent. This means that emaciating a unit's DPS won't cause it to spam non-damage skills all of the time, so you'll want your units able to do some kind of damage, even if it's only 80 or so.
2. Self-preservation is top priority. People seem to misinterpret this as "healing/revival is top priority", but this is actually not true. If a unit is going to kill the Acting unit, and the Acting unit can disable or kill the enemy before the enemy can kill it (such as with a Short Charge Paralyze), it will do that instead of casting Raise on a dead ally or healing itself, if the healing will not allow it to live through the skill. If it is in Critical and there is a dead ally within range, it will heal itself instead of Raising the dead ally. If the AI heals itself, it will always Move away from enemy units, even if moving past them would place it in range to Raise a fallen ally on the following turn. If the AI heals an ally, however, it'll still be willing to move forward, as long as the recipient of the healing skill was not the casting unit. The AI does not pay attention to the Crystal Counter (3/2/1/0), but does pay attention to turn order on dead units. If it has the choice of reviving multiple units, it will go for the one that will reach 100 CT first. It will, however, prioritize both preservation of self and allies over the death of enemies, unless that enemy is the final enemy on the map and they have the DPS to kill it in a single attack. Units with the Reraise status are generally treated as lower priority when healing, and won't be revived because the AI sees wasting a turn on making a unit get up that was going to get up anyway as a waste.
3. The AI will always use both its Move and Act command if possible, with rare exception. Idle Acts will be used to cast things like Haste, Accumulate, etc. and the AI will usually Move away from an enemy unit after attacking if it attacked before Moving.
4. MP restoration is based on both the other things the AI can do at the time, and how important it feels the target's MP-requiring skills are. Powerful skills like Flare and Slow Dance will usually see the AI using skills like Chakra and Ether to replenish MP, but if the AI views the target's skills as unimportant, it won't replenish MP, regardless of how important the skill may actually be to the current situation. The same is true of some Status-healing, such as whether or not the AI decides healing Silence is worthwhile. This is part of the reason Move-MP UP (and in Arena's case, sometimes MP Restore) is such a staple on MP-using units in AI tournaments, even more than in the single player game, on top of the obvious bit of it being MP restoration without performing Actions. The AI will still usually
use skills such as Ether correctly, and whether it decides something as being worthwhile or not is at least partially based on the team its fighting. I've had situations, for example, where my units are Silenced but the enemy is immune to most of the skills in my Talk Skill set - so the AI just won't cure Silence until something changes, which can be both a blessing and a curse depending on the situation.
5. The AI views most buffs as things to apply when a unit is somewhat damaged, as a form of light-medium healing. Reraise, Protect, Shell, etc. will get applied not at full HP, but when damaged and in need of a light heal (or in need of a large heal, if one isn't available), or sometimes when the AI is idle and has no better Act to perform, though the latter is not guaranteed. Haste, Transparent, and Wall are seen as proactive buffs that the AI will try to apply when it doesn't need to kill something. In Haste's case, it knows Speed is good, so it tries to keep itself and allies Hasted when possible. In the case of Transparent and Wall, it thinks these statuses make it invisible / invincible respectively, and will use them proactively as it sees those statuses as the best means to keep the party alive. In Vanilla, applying these proactively is impossible, and in most mods, it is either also impossible, or the statuses are altered in some way to make the above no longer true, so be careful when making a team that can use one or both of these statuses proactively. (In Arena's case, the skill Carbuncle comes to mind as a proactive source of Transparent.) When buffing a unit, though, the AI is generally smart on knowing which buffs are most important. Sometimes, it'll do something odd, like add Protect over Reraise, or add Shell instead of Protect if Protect is better for the situation and it has both, but those are similar to point 1. Just make sure the AI can only access the buffs you need it to access and this behavior should be well minimized.
6. If it has a choice between two enemy units, the AI will prioritize units with the lower Current HP, decided by its raw value. This means that a 250/250 Priest will be targeted over a 300/450 Knight, because its Current HP or 250 is less than the Knight's Current HP of 450, despite the fact the Knight is technically at "less" HP in terms of a percent (100% HP Priest v 66.6% HP Knight.) Going back to my point on keeping revival on units least likely to die, keep this in mind. You probably don't want your primary revival on units with low Max HP unless they have something like Defense UP to compensate. You can also use this trait of the AI offensively - if you want the enemy AI to ignore certain units on your squad, make them have higher Max HP and have it so that both that unit and a unit with lower Max HP enters the enemy combat range at roughly the same time.
7. There are several things the AI cannot understand properly. Firstly, it cannot properly compute Haste when it calculates the turn order, in respect to CT-bearing Spells like Flare. This is another large part as to why Haste is so gamebreaking in single-player. It also cannot properly compute when the Lancer's Jump skill will land, meaning it'll sometimes kill-steal from an allied Jumping unit. It also cannot tell what Reactions a unit has equipped and what the result of triggering said Reaction would be. For someone whose played through Vanilla FFT, this is probably the one thing on this list they've noticed if they've noticed nothing else in this entire Basic AI-Behavior section. If I've forgotten one or two things, I'll edit them in later, but these three are the most important.