Download
1 variant available
A versatile anime style LoRA with serious range. Confident linework, dramatic lighting, and color that swings from richly saturated to moody and muted depending on what you ask for. (Occult, ritual, and horror-adjacent imagery is where it's most confident and least effort, but it holds up just as well outside that lane: clean portraits, landscapes, dynamic action, multi-character scenes and NSFW).
FOR ANIMA/KREA 2 INFO, PLEASE SEE BELOW:
Trigger word: pallwash (Anima = pallwash, @pallwash)
must come before quality tags (masterpiece, best quality, etc.), not after. I find that using LESS quality tags = better result. Usually I stick with masterpiece, absurdres and that's it.
Heads up: left to its own devices on a sparse prompt, Pall Wash leans surreal and uncanny. Skulls, glowing eyes, unsettling accents show up even when you didn't ask. Want something cleaner? Be specific in your positive prompt and add a few exclusionary negatives (skull, flames, horror, undead). If you want the weird, it's as easy as hitting run.
Suggested Settings:
DPM++ 2M Karras, or Euler A / Beta 57/Karras (or experiment! It's a matter of preference)
CFG 6–6.5, (if using DaSiWa Checkpoints, lower this to 5.0)
28–30 steps.
Strength - 1.0
On checkpoints with a strong house style of their own, push weight to 1.2–1.3 (most of the time this isn't needed).
-----------------------------------------------------------------------------------------------------
Anima Version:
Recommended WF: DaSiWa Anima Workflow
Recommended Base Model: DaSiWa Obsidian Archives V2
Anima Turbo LoRA
---------
Recommended Strength: .85 - 1.0
Triggerwords: pallwash, @pallwash
CFG 4-6
Euler/linear_quadratic | er_sde/simple
40-50 Steps
Recommended Negative (for 2D anime style):
worst quality, low quality, score_1, score_2, score_3, too many fingers, deformed hand, too many arms, ugly, photorealistic, photorealism, 3D, glossy, render
---------
KREA 2:
Triggers: pallwash, @pallwash (unfortunately you need both due to my error)
Checkpoint: DaSiWa Krea2 Turbo MirroredSkies int8 (only tested on turbo lora, adjust as needed)
CFG Scale: 1 (adjust if not using a turbo checkpoint)
Steps: 8 (adjust if needed)
Sampler: euler / simple
All gallery images were ripped straight from generation. No post processing, upscaling or refining.
an example of how unguided prompts turn out as opposed to guided. (I use "Sic Mundus Creatus Est" as a wildcard because why not) (ILLUSTRIOUS)