Add anime models

cuddlehoney_v10.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:000989c0d9b089aca245e6c14a293afc34d0e33bd8acdf84c7121293732773e5
+size 2400040752

hyperfusionFinetune1_v8.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99ee86c96b25cdeab0058b67cf4bed1e2817cf61054af8a4fecf12f62fd6da36
+size 2132625888

hyperfusionFinetune1_v8.txt

@@ -0,0 +1,142 @@
+Info:
+	This model contains ~1.4 million images with a focus on hyper sized breasts/belly/ass/thighs/hips/ and more.  It is able to produce a wide variety of body shapes and sizes so experimentation is key with this model.  It's about 60% aniume, and 40% furry.  The dataset was built from a mixture of r34, e621, Chounyuu, Furaffinity, Sankaku, Gelbooru, Aryion, and Danbooru images. You can use most tags from any of these sites.  It's basically a full finetune but on multiple tagging schemes.  See the tags.csv for the full list of tags.  It really does a lot of different things now.   Starting with v8 the base model is SD1.5 instead of NovelAI, so keep that in mind if you find some tags seem to behave differently (like no more "masterpiece, best quality").
+ 
+
+Tips:
+	-artist tags included in v8 are the best way to chooose a style for the image.  Search "(artist)" in the tags csv and try them out.
+	-Negatives I commonly use with this model: "low rating, lowres, text, signature, watermark, username, blurry, transparent background, ugly, sketch, unfinished, artwork \(traditional\), multiple views"
+	-Having trouble creating prompts?  The best thing to do is copy existing tags from any booru site to see how this model reacts.  This is the best method of finding good new combinations that I have found. There is an extension to do just that for A1111 if you do some searching.
+	-Tags with less than 300 images in the dataset may not work well unless the base model already had an understanding of them, or the concept is simple enough.  More complex concepts will require 1000+ tags.
+	-If a particular tag is not working well, find similar tags to append.  The more similar tags you add, the more likley you will get what you want
+	-This model loves lots of tags.  Longer prompt lengths typically generate better quiality images.  This might be due to the training data having on average 250 words per image. Will try to increase tag dropout in the future.
+    -Remember that image orientation plays a large role in body shapes.  Wide images leave more room for a character to be wider.  Dont expect a portrait image to do hyper sizes better than square or landscape images.
+   
+   
+   
+********************************************************************************************************
+**Important Hyperfusion Tags****************************************************************************
+
+
+Body Type Tags:  This defines the overall body shape, and is a good foundation for the tags below this
+	hourglass
+	bottomheavy		
+	topheavy
+	bellyheavy   (similar to topheavy, but for the belly, mostly hyper pregnancy, with smaller breasts in comparison)
+	frontheavy   (similar to topheavy, but for big bellies AND breasts at the same time)
+		
+		
+Body Size Tags:
+	In order to unify the body size tags between the differrent sites, I trained a number of BEIT image classifiers to categorize all body sizes.  Below are the tags you should use in order from smallest to largest size. (similar to Danbooru/r-34 tags)
+	
+	Breast Size:
+    	flat chest
+        small breasts			
+        medium breasts			
+        big breasts				
+        huge breasts			
+        gigantic breasts		
+        colossal breasts		
+        
+        helper tags: topheavy, bimbo, hyper breasts
+        
+    Ass Size:
+    	medium ass			
+        big ass					
+        huge ass				
+        gigantic ass			
+        colossal ass			
+        
+        helper tags: bottomheavy, bubble butt, hyper ass, ass visible from front
+        
+    Belly Size:
+    	medium belly
+        big belly
+        huge belly
+        gigantic belly
+        colossal belly
+        
+        helper tags: bellyheavy, pregnant, hyper pregnancy
+	
+	Thigh Size: 
+		thighs
+		thick thighs
+		massive thighs
+		hyper thighs
+		
+		
+		
+Body Shape Tags: (Not as effective as the body size tags, but they work when given enough prompt weight)
+
+	Breast Shape:
+		sagging breasts
+		long breasts 		(extremely saggy, or very long, but there were not many images too train on)
+		hanging breasts 	(bent over, or leaning will help with this tag)
+		round breasts
+		breasts apart		
+		breasts on belly	(breasts resting on a big belly, or sometimes sagging to the sides of the belly)		
+		+ some other less effective shapes in tags.csv
+		
+	Ass Shape: (maybe one day I'll add some more shapes here)
+		bubble butt
+		fat ass
+		
+	Belly Shape:		
+		spherical belly		(makes a perfect sphere in shape, but not body inflation)
+		fat belly  (or obese belly)
+		chubby belly
+		stuffed belly		(this shape is characterized by the top of the belly being the same size or larger than the bottom, it works sometimes)
+		hanging belly		(bent over, or leaning position with the belly hanging down)
+		vore belly
+		sagging belly
+		torpedo belly		(belly sticking straight out like a torpedo)
+		+ some other less effective shapes in tags.csv
+		
+	Navel Shape:
+		innie navel
+		deep navel			(more extreme form of innie, but I did not classify this)
+		outie navel
+		hyper navel
+		three way navel		(for extremely fat bellies with 3 creases that intersect in the navel)
+
+	
+	
+		
+Other hyperfusion specific tags:	
+	slimthick    			(one part of the character is very skinny and the rest is not)
+	from front  			(character facing mostly forward, helps counteracting back facing shots)
+	side profile  			(character facing squarely to the side)
+	ass visible from front  (helps keep the character facing the front when using ass tags, and has a tendency towards large asses compared to "from front")	
+	backbelly  				(similar to backboob but for bellies)	
+	
+
+Other classifier labeled tags: (Existing Booru tags that I wanted to function better overall)
+	skinny					(overall skinny body, but might have large individual assets like breasts, butt, belly)
+	petite					(overall skinny body, and no large assets)
+	hyper					(any body part bigger than normally possible)
+	immobile				(any body part big enough to immobilize the persoon)
+	motion lines      		(to be used as a negative, there were only 8k images tagged with this originally)
+	sound effects			(to be used as a negative)
+	3d						(tagged all 3d/rendered images with this tag, usefull as a negative as well)
+	stuck in door
+	fat						
+	breasts on *			
+	belly on *
+	pregnancy stretch marks
+	veiny belly
+	belly blush
+	linea nigra
+	fetal movement
+	low navel				(navel placed low on the belly)
+	
+	
+	
+	
+	
+		
+**************************************************************************************************
+**All Tags****************************************************************************************
+
+	See the tag.csv file for the full list of tags.  This CSV can also be used in the A1111 autocomplete extension.  
+		This model was not trained with underscores, they are only needed in the CSV for the autocomplete extension.
+
+

hyperfusionFinetune_v7.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e196d490e92f95ee83e9a335d85068b9970a4539aeaa3e058908c4c5b686275
+size 2299933062