[IL] Final Fantasy Dawn of Souls Style (Gen Kobayashi) / ファイナルファンタジーI・IIアドバンス作風 (小林 元)
LORA
Illustrious
LORA
Illustrious
v1.2

[IL] Final Fantasy Dawn of Souls Style (Gen Kobayashi) / ファイナルファンタジーI・IIアドバンス作風 (小林 元)

⭐ 0.0
⬇ 589 Downloads
👁 1 Views
🖼 25 Images

Trained Words

F_F_D_O_S
chibi only

About this model

Attempts to recreate Gen Kobayashi’s artwork for the Final Fantasy 1+2 remake on the GBA, released in the west with the subtitle Dawn of Souls. I thought the style was neat (at least for FF1's simple, "vintage fantasy" designs. The parts of FF2's data that are more "overdesigned" don't look nearly as nice in it) and it seemed a decent candidate to try making.

Tags: F_F_D_O_S,
Optional: chibi only (for humans. It's optional for monsters.)
Aspect Ratio: The world is Square (portrait tends to stretch characters and give them longer legs, thus ruining the chibi)
The "chibi" tag is bugged in Illustrious and Pony to spawn copies of characters. Not sure how to fix it.

Trained off 36 transparent background images from Final Fantasy wiki’s FF1 and FF2 concept art pages, plus a pair of promo images of Chaos and Emperor cropped to face to avoid text and reused graphics. Flip augmentation used. Chose to do this primarily to test data needed for a style with limited scope (in this case, only character/monster images on a simple background). Works well, though wrangling with chibi's multiple views problem is a hassle (in fact, it's so much I'm not going to try the FFV chibi art I was considering doing as a followup unless someone can suggest how to avoid it).

Update: Version 1.2 changes all tags of "chibi" in training data to "chibi only" to try to clamp down on the multiple spawns thing and makes no other changes. It looks like it helps, but doesn't totally fix it.

Related Models

Similar AI models you may like