Update README.md

README.md

@@ -1,3 +1,27 @@
 ---
 license: apache-2.0
----
+---
+This LoRa is aimed at creating 3DCG female characters in indoor spaces.  With focus on elbow , neck , face , shoulder and hip areas. 
+//--//
+With focus on: Candid movement (face close ups , arm / leg movement , using 3DCG characters from Drills3D and KawaiiDetectiveEnthusiast )
+Interior design (furniture / blur / depth)
+Skin tone (3DCG - suitable for posting on regular SFW as well as NSFW)
+Liminal spaces (Pinterest photos)
+Manga frames (Claymore and Tsutomo Nihei Blame! mixed)
+Photos on adult performers at award events (safe for work! )  by morbidthoughts on flickr. 
+//--//
+This LoRa also helps limitations I have discovered with Chroma. Such as the  'cartoony'  kind of 3D aesthetic ,  in addition to struggles with scifi aesthetic.  I blame the extensive furry training in Chroma. 
+//--//
+Training data are still frames from 3DCG animation into 4x1 sliced 1024x1024 composites.
+//--//
+Ethics: Majority of training data is synthetic content.  This LoRa cannot re-create the likeness of any real world individual.  All photographic trained content are SFW sourced Pinterest photos and Flickr photos.  Training data and selection method are provided in this repository for transparency. 
+//--//
+ If you wish to prepare training sets in similiar manner the jupyter colab files have been provided in this repo. 
+//--//
+250 image composites x 4 repeats x 40 epochs
+Rank 32 Alpha 32 at LR= 0.0035 using cosine with repeats
+No bucketing , using Kohya trainer
+//---//
+Use t5xxl_fp16.safetensors text encoder or similiar if using long prompts in Chroma.
+//--//
+This LoRa is hosted on Tensor Art and SeeArt by me (Adcom).  /Cheers!