Update README.md
Browse files
README.md
CHANGED
|
@@ -25,14 +25,17 @@ Prompt: `BoKanade casting magic, 4K wallpaper, outdoors`
|
|
| 25 |
|
| 26 |
### Usage
|
| 27 |
|
| 28 |
-
The model is shared in both diffuser safetensors format.
|
| 29 |
|
| 30 |
### Concepts
|
| 31 |
|
| 32 |
The 33 concepts are listed in `concept_list` and demonstrated below.
|
| 33 |
|
|
|
|
| 34 |
|
|
|
|
| 35 |
|
|
|
|
| 36 |
|
| 37 |
|
| 38 |
|
|
@@ -40,17 +43,17 @@ The 33 concepts are listed in `concept_list` and demonstrated below.
|
|
| 40 |
|
| 41 |
|
| 42 |
|
| 43 |
-
Expect bad results for `BoMaple sheep form` and non-human concepts. Espeically the model clearly does not understand the anatomy of syrup.
|
| 44 |
-
For `BoKasumi sarashi` adding `bandages` seems to help. For `BoMaple pajama` you can add `stripe` for more similarity to the
|
| 45 |
The remaining concepts should go through smoothly.
|
| 46 |
|
| 47 |
#### Prompt format
|
| 48 |
-
During training the concept names are put at the beginning of the
|
| 49 |
Put `aniscreen` after the concept names would reinfoce the anime style.
|
| 50 |
-
|
| 51 |
-
However expect weird blending to happen most of the time starting from three concepts.
|
| 52 |
This is partially because this model is not trained too much on multi-concept scenes.
|
| 53 |
-
Below is roughly the best we can get after multiply tries (there is still
|
| 54 |
|
| 55 |
Prompt: `(BoMaple black armor) BoSally turtleneck BoKasumi, 3girls, 4K wallpaper, ahoge, black hair, brown hair, outdoors, long hair`
|
| 56 |
|
|
@@ -58,19 +61,19 @@ Prompt: `(BoMaple black armor) BoSally turtleneck BoKasumi, 3girls, 4K wallpaper
|
|
| 58 |
|
| 59 |
### More Generations
|
| 60 |
|
| 61 |
-
Prompt: BoMaple black armors aniscreen, 1girl solo, Hydra in the sky, light purple eyes, 4K wallpaper
|
| 62 |
|
| 63 |
|
| 64 |
-
Prompt: BoMaple black armors near small turtle syrup, sitting with knees up on rock looking at viewer, turtle shell, beautiful hand in glove, in front of trees , outdoors, close-up, 4K wallpaper
|
| 65 |
|
| 66 |
|
| 67 |
-
Prompt: BoMaple pajama stripe, sitting on bed with barefoot, in girl's room, detailed and fancy background, sparkling purple eyes, hand on bed, 4K wallpaper
|
| 68 |
|
| 69 |
|
| 70 |
-
Prompt: BoFrederica, cowboy shot, in rubble ruins, ((under blue sky)), cinematic angle, dynamic pose, oblique angle, 4K wallpaer, anime coloring, official art
|
| 71 |
|
| 72 |
|
| 73 |
-
Prompt: Turtle Syrup Fox Oboro next to each other simple background white background, animals
|
| 74 |
|
| 75 |
|
| 76 |
Failures are of course unavoidable
|
|
@@ -91,12 +94,12 @@ It contains 27031 images with the following composition
|
|
| 91 |
- 19279 regularization images which intend to be as various as possible while being in anime style (i.e. no photorealistic image is used)
|
| 92 |
|
| 93 |
Note that the model is trained with a specific weighting scheme to balance between different concepts so that every image does not weight equally.
|
| 94 |
-
After applying the per-image repeat we get around
|
| 95 |
|
| 96 |
|
| 97 |
### Training
|
| 98 |
|
| 99 |
-
Training is done with [EveryDream2](https://github.com/victorchall/EveryDream2trainer) trainer
|
| 100 |
I use the following configuration thanks to the suggestion of 金Goldkoron
|
| 101 |
|
| 102 |
- resolution 512
|
|
|
|
| 25 |
|
| 26 |
### Usage
|
| 27 |
|
| 28 |
+
The model is shared in both diffuser and safetensors format. Intermediate checkpoints are also shared in ckpt format in the directory `checkpoints`.
|
| 29 |
|
| 30 |
### Concepts
|
| 31 |
|
| 32 |
The 33 concepts are listed in `concept_list` and demonstrated below.
|
| 33 |
|
| 34 |
+
BoMaple +
|
| 35 |
|
| 36 |
+
BoSally +
|
| 37 |
|
| 38 |
+
The following use the full name of the concepts
|
| 39 |
|
| 40 |
|
| 41 |
|
|
|
|
| 43 |
|
| 44 |
|
| 45 |
|
| 46 |
+
Expect bad results for `BoMaple sheep form` and non-human concepts. Espeically the model clearly does not understand the anatomy and the size of syrup (well it is true it's size is not fixed).
|
| 47 |
+
For `BoKasumi sarashi` adding `bandages` seems to help. For `BoMaple pajama` you can add `stripe` for more similarity to the pajama appearing in anime.
|
| 48 |
The remaining concepts should go through smoothly.
|
| 49 |
|
| 50 |
#### Prompt format
|
| 51 |
+
During training the concept names are put at the beginning of the images separated only by spaces, but not doing so seems to work as well.
|
| 52 |
Put `aniscreen` after the concept names would reinfoce the anime style.
|
| 53 |
+
Having two concepts in a same image is fairly doable as demonstrated above.
|
| 54 |
+
However, expect weird blending to happen most of the time starting from three concepts.
|
| 55 |
This is partially because this model is not trained too much on multi-concept scenes.
|
| 56 |
+
Below is roughly the best we can get after multiply tries (there is still clothes blending).
|
| 57 |
|
| 58 |
Prompt: `(BoMaple black armor) BoSally turtleneck BoKasumi, 3girls, 4K wallpaper, ahoge, black hair, brown hair, outdoors, long hair`
|
| 59 |
|
|
|
|
| 61 |
|
| 62 |
### More Generations
|
| 63 |
|
| 64 |
+
Prompt: `BoMaple black armors aniscreen, 1girl solo, Hydra in the sky, light purple eyes, 4K wallpaper`
|
| 65 |
|
| 66 |
|
| 67 |
+
Prompt: `BoMaple black armors near small turtle syrup, sitting with knees up on rock looking at viewer, turtle shell, beautiful hand in glove, in front of trees , outdoors, close-up, 4K wallpaper`
|
| 68 |
|
| 69 |
|
| 70 |
+
Prompt: `BoMaple pajama stripe, sitting on bed with barefoot, in girl's room, detailed and fancy background, sparkling purple eyes, hand on bed, 4K wallpaper`
|
| 71 |
|
| 72 |
|
| 73 |
+
Prompt: `BoFrederica, cowboy shot, in rubble ruins, ((under blue sky)), cinematic angle, dynamic pose, oblique angle, 4K wallpaer, anime coloring, official art`
|
| 74 |
|
| 75 |
|
| 76 |
+
Prompt: `Turtle Syrup Fox Oboro next to each other simple background white background, animals`
|
| 77 |
|
| 78 |
|
| 79 |
Failures are of course unavoidable
|
|
|
|
| 94 |
- 19279 regularization images which intend to be as various as possible while being in anime style (i.e. no photorealistic image is used)
|
| 95 |
|
| 96 |
Note that the model is trained with a specific weighting scheme to balance between different concepts so that every image does not weight equally.
|
| 97 |
+
After applying the per-image repeat we get around 200K images per epoch.
|
| 98 |
|
| 99 |
|
| 100 |
### Training
|
| 101 |
|
| 102 |
+
Training is done with [EveryDream2](https://github.com/victorchall/EveryDream2trainer) trainer with [ACertainty](https://huggingface.co/JosephusCheung/ACertainty) as base model.
|
| 103 |
I use the following configuration thanks to the suggestion of 金Goldkoron
|
| 104 |
|
| 105 |
- resolution 512
|