Disclaimer

Financial support is very much appreciated, but please do so by donating to my Ko-Fi instead of sending me Buzz, as I currently have no uses for CivitAI Buzz!

Description

An SDXL 1.0 LoRa made by AI_Characters for producing images with an artstyle similar to screencaps from the anime movie Nausicaä Of The Valley Of The Wind made by Studio Ghibli - with the art primarily created by Hayao Myazaki.

Recommended generation parameters

Weight: 1.0 (adjust down as needed when used in conjunction with other LoRa's)

Model: SDXL 1.0 with 0.9 vae

Steps/Sampler/CFG/Resolution: 25 steps | DPM++ 2M Karras | CFG 7 | 1536x1536

'Kohya Deep Shrink' settings: 4 blocks | 2 downscale | 0 start | 0.35 end | skip = true

Positive prompt: high-definition 8k

Negative prompt: text, blur, low-resolution, low-quality, grainy, glossy, poorly drawn, amateur, poorly shot, compression artifacts

'CLIPTextEncodeSDXL' ComfyUI node resolution: 4096x4096

Trigger words (must be included in the prompt for the LoRa to take effect!)

• 'anime screencap in vwst artstyle'

Further usage notes

/

Usage permissions

You may not sell this model, merges containing it, or images generated with it.

Credit me when using this model in a merge or hosting it on a generation service.

TL:DR Training Workflow

1. Train full finetune with a dataset of 20-100 images using this config on a Vast.AI rented A100 80gb GPU

2. Extract LoRa at 512/512 dimensions using these settings

3. Resize LoRa to 16 ranks using these settings

Changelog

v1.2 [FINAL]

• complete rework; higher epoch count, trained with EMA, reworked captions, reworked dataset preparation - overall much better likeness and quality at the cost of some slight bias and overtraining increase - final version for now

v1.1

• retrained with new captions: slightly better style likeness, slightly better flexibility, less bias towards the subject matter of the training data

v1.0

• Initial release