khanon commited on
Commit
0ae9a7a
1 Parent(s): 96a7731

adds retrained Chise LoRA

Browse files
README.md CHANGED
@@ -18,7 +18,9 @@ Here you will find the various LoRAs I've trained, typically of Blue Archive cha
18
  [Available on old Mega.co.nz repository.](https://mega.nz/folder/SqYwQTRI#GN2SmGTBsV6S4q-L-V4VeA)
19
 
20
  ### Chise
21
- [Available on old Mega.co.nz repository.](https://mega.nz/folder/SqYwQTRI#GN2SmGTBsV6S4q-L-V4VeA)
 
 
22
 
23
  ### Hibiki
24
  [Available on old Mega.co.nz repository.](https://mega.nz/folder/SqYwQTRI#GN2SmGTBsV6S4q-L-V4VeA)
 
18
  [Available on old Mega.co.nz repository.](https://mega.nz/folder/SqYwQTRI#GN2SmGTBsV6S4q-L-V4VeA)
19
 
20
  ### Chise
21
+ [Waraku Chise / 和楽チセ / 와라쿠 치세 / 和楽千世](https://huggingface.co/khanon/lora-training/blob/main/chise/README.md)
22
+
23
+ [![Chise](chise/chara-chise-v2.png)](https://huggingface.co/khanon/lora-training/blob/main/chise/README.md)
24
 
25
  ### Hibiki
26
  [Available on old Mega.co.nz repository.](https://mega.nz/folder/SqYwQTRI#GN2SmGTBsV6S4q-L-V4VeA)
chise/README.md CHANGED
@@ -1,31 +1,55 @@
1
  # Waraku Chise (Blue Archive)
2
- Came out pretty well I think. Smaller dataset than Mari, but otherwise very similar settings.
3
 
4
- ## Usage
5
- Use any or all of these tags to summon Chise:
6
- `chise, halo, red eyes, blue hair`
7
- Hair and eyes are mostly optional if you describe a bit of her outfit as well.
8
- She naturally likes to make her `:o` expression because most art features her doing that. However I also included images tagged with other expressions.
9
- Use `open mouth`, `closed mouth`, and `parted lips` as necessary to get her to make whatever expressions you want.
 
 
10
 
11
- For her normal outfit (add as many as necessary):
12
- `braid, japanese clothes, detached sleeves, obi, tabi, geta, sailor collar, blue bow`
13
- Her shoes are weird but they're tagged `geta` and the socks as `tabi`.
 
 
 
 
 
14
 
15
- For her swimsuit outfit (add as many as necessary):
16
- `side ponytail, swimsuit, striped bikini, see-through, sailor collar, side-tie bikini bottom`
17
- You can also add `pointed ears`, which are usually only visible in her swimsuit outfit.
18
 
19
- Weight 1 works fine. Also included epoch 6 in case you find the last epoch to be a bit stubborn with her outfits.
 
20
 
21
  ## Training
22
- *All parameters are provided in the accompanying JSON files.*
23
- - Trained on a set of 119 images split by outfit, repeated 10 times (swimsuit) or 14 times (uniform) for 7 epochs (119 images * 10 or 14 repeats / 3 batch size * 7 epochs = 3178 steps)
24
- - Dataset included a mixture of SFW and NSFW.
25
- - Initially tagged with WD1.4 VITv2 model, then performed heavy pruning and editing.
26
- - Pruned implicit (`oni horns`) or redundant tags and simplified outfits so that they were always tagged with the same handful of tags
27
- - Made sure important traits were present and consitently described, and traits like `halo` were consistent with actual visibility
28
- - Added many facial expression, camera angle, and image composition hints
29
- - Used network_dimension 128 (same as usual) and network_alpha 64 (new)
30
- - This relies on the new alpha
 
 
 
 
 
31
  - Trained without VAE.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  # Waraku Chise (Blue Archive)
2
+ 和楽チセ (ブルーアーカイブ) / 와라쿠 치세 (블루 아카이브) / 和楽千世 (碧蓝档案)
3
 
4
+ [**Download here.**](chara-chise-v2.safetensors)
5
+
6
+ ## Table of Contents
7
+ - [Preview](#preview)
8
+ - [Usage](#usage)
9
+ - [Training](#training)
10
+ - [Revisions](#revisions)
11
+ - [Alternate Version](#alternate-version---v1)
12
 
13
+ ## Preview
14
+ ![Chise portrait](chara-chise-v2.png)
15
+ ![Chise moon](example-001-Defmix-v2.png)
16
+ ![Chise calligraphy](example-002-Defmix-v2.png)
17
+ ![Chise swimsuit](example-003-Defmix-v2.png)
18
+
19
+ ## Usage
20
+ Use any or all of these tags to summon Chise: `chise, halo, oni horns, hair flower, red eyes, blue hair`
21
 
22
+ For her normal outfit: `braid, japanese clothes, detached sleeves, bridal gauntlets, obi, tabi, geta, sailor collar, blue bow`
 
 
23
 
24
+ For her swimsuit alt: `side ponytail, swimsuit, striped bikini, see-through, see-through shirt, white sailor collar, side-tie bikini bottom`
25
+ - You may also add `pointy ears`, which are usually only visible in her swimsuit outfit.
26
 
27
  ## Training
28
+ *Exact parameters are provided in the accompanying JSON files.*
29
+ - Trained on a set of 119 images; 76 swimsuit, 43 normal.
30
+ - 15 repeats for normal outfit
31
+ - 9 repeats for swimsuit outfit
32
+ - 3 batch size, 4 epochs
33
+ - `(76*9 + 43*15) / 3 * 4` = 1772 steps
34
+ - 832x832 training resolution
35
+ - `constant_with_warmup` scheduler
36
+ - Initially tagged with WD1.4 Convnext-v2 model, then heavily edited.
37
+ - Pruned redundant tags and simplified outfit tags such that Chise's outfit was always described with the same words
38
+ - Ensured unique traits such as `halo` were consistent with actual visibility
39
+ - Added many tags for facial expressions, camera angles, and image composition
40
+ - Used network_dimension 128 (same as usual) / network alpha 128 (default)
41
+ - Resized to 32dim after training
42
  - Trained without VAE.
43
+
44
+ ## Revisions
45
+ - v5b (2023-02-12)
46
+ - Completely re-trained to correct issue where caption files were not being loaded correctly. New version should have more flexibility but requires additional outfit prompting.
47
+ - v1 (2023-01-26)
48
+ - Initial release.
49
+ - **Issue:** caption files were completely ignored during training. As a result, prompting "chise" grants a very strong effect, but the LoRA is highly overfit.
50
+
51
+ ## Alternate Version - c1
52
+ **v1** is also available. This version was improperly configured and ignored caption files, so it was trained on the word `chise` only. It looks quite good, however the model is very overfit and it is difficult to adjust Chise's outfit.
53
+
54
+ **[Download v1.](v1/chara-chise-v1.safetensors)**
55
+ ![Chise v1 Preview](v1/chara-chise-v1.png)
chise/chara-chise-v2.png ADDED

Git LFS Details

  • SHA256: 84364ad27a44e67c85ecf4d8bdf31ad8e8f5ee3b426ca1d7a9e6ab3ed4bbe867
  • Pointer size: 131 Bytes
  • Size of remote file: 484 kB
chise/chara-chise-v2.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5f36d3ea331c2f817acabeced642c9298ad6422aea6db8cace08682d06541307
3
+ size 37873407
chise/example-001-Defmix-v2.png ADDED

Git LFS Details

  • SHA256: 856b034226c5bf8c8ff01b319df7e8dad736c9fb59e91535a32f72636760427c
  • Pointer size: 132 Bytes
  • Size of remote file: 1.9 MB
chise/example-002-Defmix-v2.png ADDED

Git LFS Details

  • SHA256: 51522665f89117ab6440dc79fa53e575c2abc3e6e4866372fd580223bb954ddb
  • Pointer size: 132 Bytes
  • Size of remote file: 1.9 MB
chise/example-003-Defmix-v2.png ADDED

Git LFS Details

  • SHA256: 1577cad52e2e5312842247a1d04811ee874094111ba5dfb796175a16e6c2e0d9
  • Pointer size: 132 Bytes
  • Size of remote file: 1.88 MB
chise/lora_chara_chise_v2_76i9r-43i15r.json ADDED
@@ -0,0 +1,59 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "pretrained_model_name_or_path": "G:/sd/repo/models/Stable-diffusion/nai-animefull-final-pruned.safetensors",
3
+ "v2": false,
4
+ "v_parameterization": false,
5
+ "logging_dir": "",
6
+ "train_data_dir": "G:/sd/training/datasets/chise/dataset",
7
+ "reg_data_dir": "",
8
+ "output_dir": "G:/sd/lora/trained/chara/chise",
9
+ "max_resolution": "832,832",
10
+ "learning_rate": "1e-5",
11
+ "lr_scheduler": "constant_with_warmup",
12
+ "lr_warmup": "5",
13
+ "train_batch_size": 3,
14
+ "epoch": "4",
15
+ "save_every_n_epochs": "",
16
+ "mixed_precision": "fp16",
17
+ "save_precision": "fp16",
18
+ "seed": "31337",
19
+ "num_cpu_threads_per_process": 32,
20
+ "cache_latents": true,
21
+ "caption_extension": ".txt",
22
+ "enable_bucket": true,
23
+ "gradient_checkpointing": false,
24
+ "full_fp16": false,
25
+ "no_token_padding": false,
26
+ "stop_text_encoder_training": 0,
27
+ "use_8bit_adam": true,
28
+ "xformers": true,
29
+ "save_model_as": "safetensors",
30
+ "shuffle_caption": true,
31
+ "save_state": false,
32
+ "resume": "",
33
+ "prior_loss_weight": 1.0,
34
+ "text_encoder_lr": "1.5e-5",
35
+ "unet_lr": "1.5e-4",
36
+ "network_dim": 128,
37
+ "lora_network_weights": "",
38
+ "color_aug": false,
39
+ "flip_aug": false,
40
+ "clip_skip": 2,
41
+ "gradient_accumulation_steps": 1.0,
42
+ "mem_eff_attn": false,
43
+ "output_name": "chara-chise-v2",
44
+ "model_list": "",
45
+ "max_token_length": "150",
46
+ "max_train_epochs": "",
47
+ "max_data_loader_n_workers": "",
48
+ "network_alpha": 128,
49
+ "training_comment": "Character: `chise, halo, red eyes, blue hair, oni`\nStandard outfit: `braid, japanese clothes, detached sleeves, obi, tabi, geta, sailor collar, blue bow`\nSwimsuit: `side ponytail, swimsuit, striped bikini, see-through, sailor collar, side-tie bikini bottom, pointed ears`\n(76 swimsuit * 9 repeats + 43 normal *15 repeats)/ 3 batch size * 4 epochs = 1772 steps",
50
+ "keep_tokens": 2,
51
+ "lr_scheduler_num_cycles": "",
52
+ "lr_scheduler_power": "",
53
+ "persistent_data_loader_workers": true,
54
+ "bucket_no_upscale": true,
55
+ "random_crop": false,
56
+ "bucket_reso_steps": 64.0,
57
+ "caption_dropout_every_n_epochs": 0.0,
58
+ "caption_dropout_rate": 0
59
+ }
chise/v1/chara-chise-v1.png ADDED

Git LFS Details

  • SHA256: 91abd6dc4ae0f3537611ab2eb7c332e9813a78af1828c5f1a1af8226d7e96ab6
  • Pointer size: 131 Bytes
  • Size of remote file: 535 kB
chise/v1/chara-chise-v1.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:07cc79fffeab73e4931864c5f12a36e8393f18674a2f3612b906329901a94b67
3
+ size 151110333
chise/{lora_chara_chise_v1_119i13r_832_batch3_5e-5text_2e-4unet_7epoch.json → v1/lora_chara_chise_v1_119i13r_832_batch3_5e-5text_2e-4unet_7epoch.json} RENAMED
File without changes