I was trying to train my Au Ra in FF14, by providing screenshot from the game. Its can output the horn and scales right, but the 3D style is affecting the generation. Anyone have a idea of how to fix it?These are some of the settingpretrained_model_name_or_path = "./sd-models/latest.ckpt"resolution = "640,640"enable_bucket = truemin_bucket_reso = 256max_bucket_reso = 1024save_every_n_epochs = 1max_train_epochs = 10train_batch_size = 2network_train_unet_only = falsenetwork_train_text_encoder_only = falselearning_rate = 0.0001unet_lr = 3.5e-5text_encoder_lr = 3e-6lr_scheduler = "cosine_with_restarts"optimizer_type = "Lion"lr_scheduler_num_cycles = 1network_module = "networks.lora"network_dim = 32network_alpha = 32shuffle_caption = trueweighted_captions = falsekeep_tokens = 1max_token_length = 255seed = 1337prior_loss_weight = 1clip_skip = 2gradient_checkpointing = falsemixed_precision = "fp16"save_precision = "fp16"xformers = truelowram = falsecache_latents = truecache_latents_to_disk = falsepersistent_data_loader_workers = true

Aura (FFxiv)

I fixed the repeat rate of body Data to 6, and 10 for head data. And use AOM3 as base model. Can output good horn and scale. but the 3D feel still kinda affecting the output. Also cut the trigger word to Only Ares. please put covered_ears to prevent the human ears to pop-out the hand still a issue cuz the in-game hand is large.

Files with SHA256: 5e6af2ff8a061c6de31dafe8cbc90a9a7b102c7cde7c0d29e94a345b3905916c