Based on the GentleBean outfit set that was gainable from the FALL GUYS crossover that they've done twice without me BEING ON FOR IT grumbles (I even missed the 16 event - I don't have torgal i'm really sadge.)
Activation tag: gentlebean_ffxiv
Tips:
Less 3D: Use it at strengths under 0.8
More 3D like the game: 0.8 + 3d/Realistic
As of this writing, there MAY not be a re-make, because on hi-res and normal res it should be fine with the settings above. Normally I have more time to do these things but i've been on the fly making things.
Link list coming soon.
Description
FAQ
Comments (10)
Seems to be a good amount of style pollution and seems to makes very long legs. Was all of the training data game screenshots? Did you tag them "3d, game screenshot"? Pony really wants to be told any polygon based training data is 3D from what I've seen.
yes, it was data game screenshots, but that's what this was meant to be, In game -- is 3d.
I'm not saying i'm RIGHT in how i trained it btw, i'm totally willing to take tips, like some of my clothing loras are really hit or miss. I probably shoudl've added extra hoodie data or something in it.
Ugh also it could be how i messed with the dim/alpha that messed that up too, normally i'm doing 32/32 but since i mostly do styles/clothjing/.backgrounds
@duskfallcrew My limited experience with training stuff using 3D models and more extensive experience using LoRAs (where this kind of issue is obvious) is that having "3D" in the training tags really minimizes the issues with 3D training data. I expected at least most of it would be game screenshots, though I asked with italicized "all" because there might a promo piece, menu icon, piece of fan art or such thrown in. Can't help on the dim/alpha front.
@NanashiAnon Ok well, sometimes if i have 3d in the training tags it over emphasises it - lol i'm getting hit/miss on pony so i'm happy to check my dataset and try again. I know i have to re-do the archeo one on pony because i had LITTLE extra data. The other HARD part is i'm also doing like severely limited sized screenshots because the data i have is under SDXL minimum, it's doable it just is a bit not as easy.. and i forgot to add in my own game data (as in my miqo'te ) but i try not to do that part because it leans too hard into one thing
@duskfallcrew Was the overemphasis present with 3D as a negative tag in prompt?
I'm not sure if you're WANTING it to be3d or NOT 3d - i don't always tag it as 3d - but gentlebean was a hard one because i wasn't actually on sub when it was on xD noramlly like i said i have myiad of my own data to throw in but i wasn't subbed both times this was on
@duskfallcrew Tagging stuff 3D in training data helps you use 3D in negative prompt to get rid of 3Dness (if you don't use it in TD, the LoRA doesn't seem to know the image is 3D and just assumes that's how it's drawn). I'm not sure how much impact it has, but I also throw in "game screenshot" for completeness (since it's a common enough tag on boorus).
@NanashiAnon Yea that's ... not .. how i was originally just -- LOL originally you never tagged 3d because if you tagged it and it SAW it it'd do it. The thing is if anything i didn't intend for this NOT to be 3d- I'm happy to make a LESS 3d option and re-tag it? Honestly, the only downside is having to do double duty non 3d is the cost of training at nearly 1-2k a pop for how many images i'm doing.
@NanashiAnon No? If you checked the pictures, there is no "3d" in the negative. Again, i'm not trying to be rude or anything but ti's FFXIV game data, the intent wasn't to make a non 3d lora - if you look at a lot of my other XIV content for Pony the 3d is very strong even WITH 3d in the tags.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.