ベアゲルターのアリマ可愛らしさとかわいそうさの再現度は低い気変わった手法で学習させた 最初の50エポックはアリマ(Alim)が写っていれば他の人物も写ってる戦闘シーンとかの画像も使って学習 それではアリマに対して理解度が低かったのでアリマだけが写っている画像に減らして前の50エポックから更に50エポック学習させた"A unique training method was employed. The first 50 epochs utilized images of battle scenes or other contexts where Alim was present, along with other people. However, as this approach did not sufficiently enhance the understanding of Alim, the training set was narrowed down to images featuring Alim alone, and training continued for an additional 50 epochs from that point."Nega<pre><code>speech bubble, letterboxed, pillarboxed, black border, white border</code></pre>v002ベアゲルターLoRAシリーズの経験をフィードバックそれなりの出来に

アリマ-XL Alim-XL Lora (ベアゲルター/Die Wergelder)

出来が悪いとは思っていたのでその後のベアゲルターキャラの経験をフィードバックしてやり直しベースモデルもebaraに変更黒いセーラー服を着せないとアリマっぽくならないのは注意<pre><code> "ss_sd_model_name": "ebara_pony_1.bakedVAE.safetensors",
 "ss_resolution": "(1024, 1024)",
 "ss_clip_skip": "None",
 "ss_num_train_images": "745",
 "ss_enable_bucket": "True",
 "ss_lr_scheduler": "constant",
 "ss_max_token_length": "225",
 "ss_sd_scripts_commit_hash": "6b1520a46b1b6ee7c33092537dc9449d1cc4f56f",
 "ss_max_train_steps": "600",
 "ss_batch_size_per_device": "2",
 "ss_v2": "False",
 "ss_cache_latents": "True",
 "ss_mixed_precision": "bf16",
 "ss_random_crop": "False",
 "ss_gradient_accumulation_steps": "32",
 "ss_optimizer": "prodigyopt.prodigy.Prodigy(safeguard_warmup=True)",
 "ss_learning_rate": "1.0",
 "ss_max_grad_norm": "1.0",
 "modelspec.resolution": "1024x1024",
 "ss_prior_loss_weight": "0.1",
 "ss_keep_tokens": "2",
 "ss_base_model_version": "sdxl_base_v1-0",
 "ss_num_reg_images": "0",
 "ss_bucket_no_upscale": "True",
 "ss_network_dim": "32",
 "ss_caption_dropout_rate": "0.05",
 "ss_num_epochs": "50",
 "ss_num_batches_per_epoch": "380",
 "ss_network_alpha": "1.0",
 "ss_network_module": "networks.lora",
 "modelspec.date": "2024-03-31T00:04:27",
 "ss_gradient_checkpointing": "True",
 "ss_zero_terminal_snr": "False",
 "ss_face_crop_aug_range": "None",
 "modelspec.sai_model_spec": "1.0.0",
 "ss_unet_lr": "None",
 "ss_seed": "946220687",
 "ss_lr_warmup_steps": "0",
 "ss_lowram": "False",
 "ss_text_encoder_lr": "None",
 "ss_output_name": "Alim-XL_003",
 "ss_multires_noise_discount": "0.3",
 "ss_total_batch_size": "64",
 "ss_training_started_at": "1711779167.2030387",
 "ss_color_aug": "False",
 "ss_session_id": "1431016150",
 "modelspec.prediction_type": "epsilon",
 "ss_min_snr_gamma": "5.0",
 "ss_steps": "480",
 "ss_training_finished_at": "1711811067.5053518",
 "ss_min_bucket_reso": "256",
 "ss_caption_dropout_every_n_epochs": "0",
 "ss_shuffle_caption": "True",
 "ss_max_bucket_reso": "2048",
 "ss_flip_aug": "False",
 "ss_epoch": "40"</code></pre>

Files with SHA256: 29967eab3db847740215e2d611af37c2dc937aac208656a2b8ea50c0ce9eeb4b