⚔️ Nush from XTER
Nice Mommy is a NICE MOMMY 
🧠
TRAINED with Anima-Standalone-Trainer (https://github.com/gazingstars123/Anima-Standalone-Trainer)
DIT model below
https://civarchive.com/models/2414435/animax-anima-finetune-model
animaxAnimaFinetune_v05.safetensors
⚙️ Training Info
Trainer: Trained with Anima-Standalone-Trainer on an RTX 3080 10GB.
Base DiT Model: animaxAnimaFinetune_v05.safetensors
TAGS - nush, xter,
📦 Dataset & Training Details (No Lies Edition) for V1.
(V2[epoch 2/3 checkpoint] is much improved - go for that one for more detailed Nush look)
Keeping it 100% transparent about the dataset quality so you know exactly what to expect:
V1.0 details / for v2.0 check changelog notes
Dataset Size: 141 images.
Quality & Variety: Varied! The dataset contains a mix of black and white sketches, manga panels, a few screenshots from hentai, doujinshi illustrations, and fan art.
Cleanup: Manually edited to remove watermarks, text tags, or background characters from the images.
Tagging: Pre-processed and auto-tagged using Kohya_ss (WD14 captioner with Threshold 0.5 / General threshold 0.35 / Character threshold 0.35 confidence settings), then manually reviewed to add missing details and delete conflicting tags.
❤️ Share Your Generations!
Description
#### 📢 Release Notes: Version 1.0 (Public Initial Release)
While this is the very first version being shared publicly on Civitai, under the hood, this is my V2 iteration of the model! I completely overhauled my internal test version to make sure it was ready for the community.
---
### 🔄 What's New & Improved (Compared to my private V1 test):
* Dataset Expansion: Expanded the dataset from 88 to 141 images for much better flexibility and concept coverage. (Full disclosure: her signature mole under the lip is still a bit inconsistent, but the overall likeness is way stronger!)
* Deep Dataset Cleanup: Manually went through the entire dataset to meticulously remove text tags, intrusive watermarks, and background characters.
* Refined Captions: Switched to an optimized Kohya_ss WD14 setup for base tagging and manually pruned the results to prevent concept bleeding.
I'll probably try to crack that mole issue and force it to be 100% consistent for the next public update (V2), but for now, this is a massive upgrade. Enjoy!