
Always wanted to make this since DunMa and DungeonSquad are among my favorite Mobile Games, I was finally motivated by ElHuevoCosmic's Dungeon Squad Style LoRA, which was sadly trained on a faulty dataset.
That being said mine is not entirely without issues either:
-
Gradient Backgrounds are a little too baked in (I went a little overboard to compensate for Cosmic's issue. Will fix in next version.)
-
Random weapons appear
-
Pose hard to influence (will fix)
However no green boarder issues and if you want better success at backgrounds use something like tPonyNaiV6 (Pony itself is bad at backgrounds).
It's important that you don't use quality enhancing embeddings like UnaestheticXL negative or zPDXL2 as these tend to force smoother shading, removing the pixel art effect from shading.
Use the amazing Void91's HDA PixelArtXL embedding for best results.
Comes in 3 flavors:
Regular - Good results, weak backgrounds.
Strong - Overtrained, better at backgrounds, works best when used with "pixel_art" in negative to make smooth looking characters in the Dungeon Squad anatomy/pose/framing.
Weak - good for img2img upscaling of pixel art gens.
描述:
General purpose
训练词语: pixel_art
名称: DungeonSquadStylePony-Regular.safetensors
大小 (KB): 111769
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success