onimai

grasswonder-umamusume/README.md

@@ -35,7 +35,7 @@ Trained with [Kohya trainer](https://github.com/Linaqruf/kohya-trainer)
 ![native-00025-570458801](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/grasswonder-umamusume/samples/native-00025-570458801.png)
 
 
-### LoRA embedding
+### LoRA
 
 Please refer to [LoRA Training Guide](https://rentry.org/lora_train)
 
@@ -44,7 +44,7 @@ Please refer to [LoRA Training Guide](https://rentry.org/lora_train)
 - learning rate 1e-4
 - batch size 6
 - clip skip 2
-- number of training steps 7520 (20 epochs)
+- number of training steps 7520/6 (20 epochs)
 
 *Examples*
 ![lora-00014-2366006784](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/grasswonder-umamusume/samples/lora-00014-2366006784.png)

onimai/README.md

@@ -0,0 +1,67 @@
+This folder contains models trained for the two characters oyama mahiro and oyama mihari.
+
+Trigger words are
+- oyama mahiro
+- oyama mihari
+
+To get anime style you can add `aniscreen`
+
+At this point I feel like having oyama in the trigger is probably a bad idea because it seems to cause more character blending.
+
+
+### Dataset
+
+Total size 338
+
+screenshots 127
+- Mahiro: 51
+- Mihari: 46
+- Mahiro + Mihari: 30
+
+fanart 92
+- Mahiro: 68
+- Mihari: 8
+- Mahiro + Mihari: 16
+
+Regularization 119
+
+For training the following repeat is used
+- 1 for Mahiro and reg
+- 2 for Mihari
+- 4 for Mahiro + Mihari
+
+
+### Base model
+
+[NMFSAN](https://huggingface.co/Crosstyan/BPModel/blob/main/NMFSAN/README.md)
+
+
+### LoRA
+
+Please refer to [LoRA Training Guide](https://rentry.org/lora_train)
+
+- training of text encoder turned on
+- network dimension 64
+- learning rate scheduler constant
+- learning rate 1e-4 and 1e-5 (two separate runs)
+- batch size 7
+- clip skip 2
+- number of training epochs 45
+
+
+### Comparaison
+
+learning rate 1e-4
+
+![grid-00010-492069042](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/onimai/samples/grid-00010-492069042.png)
+
+learning rate 1e-5
+![grid-00017-492069042](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/onimai/samples/grid-00017-492069042.png)
+
+Normally with 2 repeats and 45 epochs we should have perfectly learned the character with dreambooth (using typically lr=1e-6), but here with lr=1e-5 it does not seem to work very well. lr=1e-4 produces quite correct results but there is a risk of overfitting.
+
+### Examples
+
+![00026-4010692159](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/onimai/samples/00026-4010692159.png)
+![00030-286171376](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/onimai/samples/00030-286171376.png)
+![00034-2431887953](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/onimai/samples/00034-2431887953.png)

suremio-nozomizo-eilanya-maplesally/README.md

@@ -44,6 +44,7 @@ Regularization 276
 
 [NMFSAN](https://huggingface.co/Crosstyan/BPModel/blob/main/NMFSAN/README.md) so you can have different styles
 
+
 ### Native training
 
 Trained with [Kohya trainer](https://github.com/Linaqruf/kohya-trainer)
@@ -60,7 +61,9 @@ Trained with [Kohya trainer](https://github.com/Linaqruf/kohya-trainer)
 ![native-00010-2248582025](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/suremio-nozomizo-eilanya-maplesally/samples/native-00010-2248582025.png)
 ![native-00014-3296158149](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/suremio-nozomizo-eilanya-maplesally/samples/native-00014-3296158149.png)
 ![native-00048-3129463315](https://huggingface.co/alea31415/YuriDiffusion/resolve/main/suremio-nozomizo-eilanya-maplesally/samples/native-00048-3129463315.png)
-### LoRA embedding
+
+
+### LoRA
 
 Please refer to [LoRA Training Guide](https://rentry.org/lora_train)
 
@@ -69,7 +72,7 @@ Please refer to [LoRA Training Guide](https://rentry.org/lora_train)
 - learning rate 1e-4
 - batch size 6
 - clip skip 2
-- number of training steps 69700 (50 epochs)
+- number of training steps 69700/6 (50 epochs)
 
 *Examples*