khanon commited on
Commit
0896449
1 Parent(s): c696165

adds retrained Reisa LoRA

Browse files
reisa/README.md CHANGED
@@ -1,23 +1,49 @@
1
- # Reisa (Blue Archive)
2
- Came out okay, but not great, largely due to a lack of quality training data. Very overfit on her outfit so I recommend dropping the weight to 0.85 to dress her up in other stuff.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
 
4
  ## Usage
5
- Use any or all of these tags to summon Reisa:
6
- `reisa, halo, low twintails, pink hair, purple eyes, star hair ornament`
7
- Hair and eyes are mostly optional if you describe a bit of her outfit as well.
8
- She naturally likes to make her `:D` expression because most art features her doing that. However I also included images tagged with other expressions and you can get her to perform other expressions reasonably easily.
9
 
10
- For her normal outfit (she will probably show up in this by default, so negative prompt these to change her)
11
- `school uniform, grey shirt, blue jacket, pleated skirt, striped socks, backpack, sneakers`
 
 
12
 
13
- I suggest lowering the weight or using epoch 6 if she's really fighting against you changing her outfit.
14
 
15
  ## Training
16
- *All parameters are provided in the accompanying JSON files.*
17
- - Trained on a set of 89 images, repeated 17 times 7 epochs (89 images * 17 repeats / 3 batch size * 7 epochs = 3530 steps)
18
- - Dataset included a mixture of SFW and NSFW.
19
- - Initially tagged with WD1.4 Convnextv2 model, then performed heavy pruning and editing.
20
- - Made sure important traits were present and consitently described, and traits like `halo` were consistent with actual visibility
21
- - Added many facial expression, camera angle, and image composition hints
22
- - Used network_dimension 128 (same as usual) and network_alpha 64
 
 
 
23
  - Trained without VAE.
 
 
 
 
 
 
 
 
 
 
1
+ # Uzawa Reisa (Blue Archive)
2
+ 宇沢レイサ (ブルーアーカイブ) / 우자와 레이사 (블루 아카이브)
3
+
4
+ [**Download here.**](https://huggingface.co/khanon/lora-training/blob/main/reisa/chara-reisa-v3.safetensors)
5
+
6
+ ## Table of Contents
7
+ - [Preview](#preview)
8
+ - [Usage](#usage)
9
+ - [Training](#training)
10
+ - [Revisions](#revisions)
11
+
12
+ ## Preview
13
+
14
+ ![Reisa portrait](chara-reisa-v3.png)
15
+ ![Reisa V](example-001-AOM2NMMElyv3.png)
16
+ ![Reisa pose](example-002-DefmixRed-v3.png)
17
+
18
+ Unfortunately the model frequently struggles with Reisa's halo.
19
 
20
  ## Usage
 
 
 
 
21
 
22
+ Use any or all of these tags to summon Reisa: `reisa, halo, low twintails, pink hair, purple eyes, star hair ornament`
23
+ - Hair and eye tags are optional.
24
+
25
+ She likes to make her `:d` expression by default. I included images tagged with other expressions, so you should be able to change it, but if not try adding `:d` to the negative prompt.
26
 
27
+ For her normal outfit: `school uniform, grey shirt, blue jacket, pleated skirt, striped socks, backpack, sneakers`
28
 
29
  ## Training
30
+ *Exact parameters are provided in the accompanying JSON files.*
31
+ - Trained on a set of 89 images
32
+ - Datset included a mixture of SFW/NSFW
33
+ - 13 repeats
34
+ - 3 batch size, 4 epochs
35
+ - `(89*13) / 3 * 4` = 1543 steps
36
+ - 768x768 training resolution
37
+ - `constant_with_warmup` scheduler instead of `cosine`
38
+ - Initially tagged with WD1.4 Convnextv2 model. Tags were edited for accuracy and improved outfit consistency. Tags for facial expressions, camera angle, and image composition were added.
39
+ - Used network_dimension 128 (same as usual) / network alpha 128 (default)
40
  - Trained without VAE.
41
+ - Training dataset available here.
42
+
43
+ ## Revisions
44
+ - v3 (2023-02-06)
45
+ - 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.
46
+ - v2b (2023-01-28)
47
+ - Initial release.
48
+ - **Issue:** caption files were completely ignored during training. As a result, prompting "reisa" grants a very strong effect, but the LoRA is highly overfit.
49
+ - [Download available on Mega.co.nz repository.](https://mega.nz/folder/yjwGiIBD#w79CxW2-Cj66QmObg0zOdg)
reisa/chara-reisa-v3.png ADDED

Git LFS Details

  • SHA256: ca41958822535ffe855e0f46f6849025c955f9b77ba0f8ac80c5e73fe4e59154
  • Pointer size: 131 Bytes
  • Size of remote file: 496 kB
reisa/chara-reisa-v3.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0ef0d52aeff5134ecf5755b6a3e59275a344d30a5bf3ef61982c2ef73099afe9
3
+ size 151118356
reisa/example-001-AOM2NMMElyv3.png ADDED

Git LFS Details

  • SHA256: 6a93c809acacd13d9be2af7505f699b6084ec7a2c9778ab73c988d4f104d71f6
  • Pointer size: 132 Bytes
  • Size of remote file: 7.75 MB
reisa/example-002-DefmixRed-v3.png ADDED

Git LFS Details

  • SHA256: c7902fb0503609a4c2b68e1ca47714f205893aa4c16bd7927387e593fea0d607
  • Pointer size: 132 Bytes
  • Size of remote file: 2.15 MB
reisa/lora_chara_reisa_v2_3530step_832_batch3_5e-5text_2e-4unet_7epoch.json DELETED
@@ -1,50 +0,0 @@
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/reisa/dataset",
7
- "reg_data_dir": "G:/sd/training/datasets/regempty",
8
- "output_dir": "G:/sd/lora/trained/chara/reisa",
9
- "max_resolution": "832,832",
10
- "learning_rate": "",
11
- "lr_scheduler": "cosine_with_restarts",
12
- "lr_warmup": "5",
13
- "train_batch_size": 3,
14
- "epoch": "7",
15
- "save_every_n_epochs": "2",
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": "",
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": "5e-5",
35
- "unet_lr": "2e-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-reisa-v2",
44
- "model_list": "",
45
- "max_token_length": "150",
46
- "max_train_epochs": "",
47
- "max_data_loader_n_workers": "",
48
- "network_alpha": 64,
49
- "training_comment": ""
50
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
reisa/{lora_chara_reisa_v1_3115step_768_batch3_5e-5text_2e-4unet_7epoch.json → lora_chara_reisa_v3_89i13r.json} RENAMED
@@ -4,21 +4,21 @@
4
  "v_parameterization": false,
5
  "logging_dir": "",
6
  "train_data_dir": "G:/sd/training/datasets/reisa/dataset",
7
- "reg_data_dir": "G:/sd/training/datasets/regempty",
8
  "output_dir": "G:/sd/lora/trained/chara/reisa",
9
  "max_resolution": "768,768",
10
- "learning_rate": "",
11
- "lr_scheduler": "cosine_with_restarts",
12
  "lr_warmup": "5",
13
  "train_batch_size": 3,
14
- "epoch": "7",
15
- "save_every_n_epochs": "2",
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": "",
22
  "enable_bucket": true,
23
  "gradient_checkpointing": false,
24
  "full_fp16": false,
@@ -31,8 +31,8 @@
31
  "save_state": false,
32
  "resume": "",
33
  "prior_loss_weight": 1.0,
34
- "text_encoder_lr": "5e-5",
35
- "unet_lr": "2e-4",
36
  "network_dim": 128,
37
  "lora_network_weights": "",
38
  "color_aug": false,
@@ -40,11 +40,15 @@
40
  "clip_skip": 2,
41
  "gradient_accumulation_steps": 1.0,
42
  "mem_eff_attn": false,
43
- "output_name": "chara-reisa-v1",
44
  "model_list": "",
45
  "max_token_length": "150",
46
  "max_train_epochs": "",
47
  "max_data_loader_n_workers": "",
48
- "network_alpha": 64,
49
- "training_comment": ""
 
 
 
 
50
  }
 
4
  "v_parameterization": false,
5
  "logging_dir": "",
6
  "train_data_dir": "G:/sd/training/datasets/reisa/dataset",
7
+ "reg_data_dir": "",
8
  "output_dir": "G:/sd/lora/trained/chara/reisa",
9
  "max_resolution": "768,768",
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,
 
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,
 
40
  "clip_skip": 2,
41
  "gradient_accumulation_steps": 1.0,
42
  "mem_eff_attn": false,
43
+ "output_name": "chara-reisa-v3",
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: `reisa, halo, low twintails, pink hair, purple eyes, star hair ornament` (hair and eyes optional)\nOutfit: `school uniform, grey shirt, blue jacket, pleated skirt, striped socks, backpack, sneakers`\n\n89 images * 13 repeats / 3 batch size * 4 epochs = 1543 steps",
50
+ "keep_tokens": 2,
51
+ "lr_scheduler_num_cycles": "",
52
+ "lr_scheduler_power": "",
53
+ "persistent_data_loader_workers": true
54
  }