all images in sample were made with the model "ANIMA BASE"
USE THE TRIGGER WORDS BELOW:
trigger_word - @rebaar.
ABSOLUTE ANIMA EDITION!!!
I use an image resolution of 960x1280
and highly recommend enabling Hires. fix,( 1.5X UPSCALE 0.3/0.4 DENOISING) as it provides impressive results compared to standard generation without it (Better style similarity). I recommend using REMARCI or any other custom upscaler of your choice.
Additionally, the LoRA strength/weight can be adjusted to 1.2, or you can choose not to use the trigger word at all."
Quality tags - masterpiece, best quality, score_9, score_8, score_7, score_6
Negative - worst quality, low quality, lowres, score_1, score_2, score_3, blurry, jpeg artifacts, sepia, early, loli, bad hands, missing fingers, artist name, watermark, cross, bad details,
ER SDE + SGM UNIFORM, SHIFT 4/3, GFG SCALE 4/3, CLIP SKIP 1/2
all images in sample also use the lora weight at 1.00 and 1.2
once again always give support to the real artists.
This model is for artistic and educational purposes only. The use of certain styles is based on publicly available data and is not intended to infringe on the intellectual property of any artists, nor is it to be made for commercial use or to earn any money. Please respect the work of original creators
Description
version with 1 fried trigger word, (another 2 trigger words works well)
FAQ
Comments (3)
Awesome style!
Works nicely thanks. But in the future I would recommend against using actual tags as triggers, since some may not want the effects of those tags in the results.
I just tried training a new, fixed LORA model on AdamW without a text encoder and using a corrected dataset and one trigger word, and after five hours, I realized it’s been overtrained into a shit, it’s producing artifacts and the quality is worse than the original. Anyway... I’ll still need to test the settings with a small dataset, but thanks for the comment anyway