banner

Akihiko Yoshida styles for Stable Diffusion

Final Fantasy Tactics, Tactics Ogre, Bravely Default, Nier Automata, etc

There is no rhyme or reason for mixing and weighing these, just have fun. You can use all of them at once and you'll get a Super Tactics style, or any of them individually. They're all great. I'll give some basic tips based on my experiences as a potential baseline for you but, really, anything goes.

Tactics Ogre LUCT Hypernetwork NAI - https://mega.nz/file/k8knyQST#5YUVf39JBybhBybSM8Ul9FPS0MGxIqtLtffFWLOXPKI
"NAI Tactics Ogre LUCT Hypernetwork. Works on NAI/Anything and their mixes so far. No named character recognition though, you have to prompt for their looks. Some samples included. Trained on the art book. Only humans."

I usually have this enabled at very low strength regardless of what I'm doing. Even 0.1 is good enough to add some of that nostalgic styling that will fight against the modern styles your model and the Akihiko Yoshida Lora will supply. Setting the strength high can create some very unusual and abstract images.

Tactics Ogre LUCT Lora NAI - https://mega.nz/folder/o4sUFAZZ#tJrV5pCTgSrTdsZiNdWcCA
"I also used the same training images from the hypernetwork. Results looks comparable to the hypernetwork."

While the author says the style is comparable to the hypernetwork, that's not what I see. It certainly looks like Tactics Ogre but to me it's more clean and modern, like the LUCT style. Whereas the hypernetwork leans towards the older style.

Akihiko Yoshida (Bravely Default, FF Tactics, NieR) Style LoRA - https://civitai.com/models/5558/akihiko-yoshida-bravely-default-ff-tactics-nier-style-lora

This really boosts the FFT style I was after but it has a lot of data from other games, one time I put "Bravely Default" into the negatives and it helped a bit with what I was going for. I tend to keep this at around 0.6.

Tactics Ogre Lora trained on the 90s character art from the artbook - https://files.catbox.moe/6vtueu.safetensors
"Gives you the armor/cloth shading, the isometric stance, and 90s Yoshida faces. I excluded the illustrations and monsters from training this version as they made the armor/clothing too bulky and plain."

I find that this one affects the background and lighting more than the others which are more about the characters. It adds an official illustration feeling.

FFT Avatars - https://www.dropbox.com/s/9v5m7oymkdum9sm/FFT.safetensors?dl=0
"I tried my best without putting in actual effort. idek what franchise this is so your triggerwords are "pixelart, portrait". also, 1boy, 1girl, blonde hair and brown hair were randomly added to the images since I couldn't be assed to tag them properly lmao."

Trained on the FFT character portraits. I found that you can weigh the "pixelart" part negatively while still getting the "intended" style of the Lora, IE (pixelart:-0.3), but eventually I just removed that part entirely.

Here are a bunch of prompts to try, download the images from this mega link.
https://mega.nz/folder/UYZBUAhR#hCRNEcZFqHiolnzcjcQTMQ
image description
image description
image description
image description
image description
image description
image description
image description
image description
image description
image description
image description
image description
image description
image description
image description

This is my twitter if you want to contact me https://twitter.com/Elf_Anon

Edit
Pub: 13 Feb 2023 07:14 UTC
Edit: 19 Apr 2023 22:42 UTC
Views: 2424