Island Generator (SDXL, FFXL) : v.0.4 SDXL - LoRA CivitAI: Island_Generator_Update.safetensors

Switching to SDXL 1.0 LoRA with CivitAI trainer

Lyco version TBA

"ss_num_train_images": "700",
"modelspec.sai_model_spec": "1.0.0",
    "ss_max_grad_norm": "1.0",
    "sshs_legacy_hash": "1b8ce68d",
    "ss_gradient_accumulation_steps": "1",
    "ss_num_epochs": "10",
    "ss_gradient_checkpointing": "True",
    "ss_scale_weight_norms": "None",
    "ss_unet_lr": "0.0005",
    "ss_num_batches_per_epoch": "179",
    "ss_output_name": "Island_Generator_Update",
    "ss_mixed_precision": "bf16",
    "ss_network_dropout": "None",
    "ss_network_alpha": "1",
    "ss_lowram": "False",
    "ss_caption_tag_dropout_rate": "0.0",
    "ss_caption_dropout_every_n_epochs": "0",
    "ss_v2": "False",
    "ss_training_comment": "None",
    "ss_session_id": "1805248317",
    "ss_training_finished_at": "1698238155.6951559",
    "ss_optimizer": "bitsandbytes.optim.adamw.AdamW8bit(weight_decay=0.1)",
    "ss_max_train_steps": "1790",
    "modelspec.resolution": "1024x1024",
    "ss_lr_warmup_steps": "0",
    "modelspec.title": "Island_Generator_Update",
    "ss_min_snr_gamma": "5.0",
    "ss_face_crop_aug_range": "None",
    "ss_cache_latents": "True",
    "ss_network_module": "networks.lora",
    "ss_dataset_dirs": {
        "img": {
            "n_repeats": 1,
            "img_count": 700
        }
    },
    "ss_new_sd_model_hash": "e6bb9ea85bbf7bf6478a7c6d18b71246f22e95d41bcdd80ed40aa212c33cfeff",
    "ss_lr_scheduler": "cosine_with_restarts",
    "ss_multires_noise_iterations": "6",
    "ss_prior_loss_weight": "1.0",
    "modelspec.implementation": "https://github.com/Stability-AI/generative-models",
    "modelspec.prediction_type": "epsilon",
    "sshs_model_hash": "f9eba922d4c46abcadf0af267051db3bba3c367cf5d8a37792ecf7527661a017",
    "ss_adaptive_noise_scale": "None",
    "ss_zero_terminal_snr": "False",
    "ss_caption_dropout_rate": "0.0",
    "ss_text_encoder_lr": "5e-05",
    "modelspec.architecture": "stable-diffusion-xl-v1-base/lora",
    "ss_network_dim": "32",
    "ss_multires_noise_discount": "0.3",
    "ss_epoch": "10",
    "ss_steps": "1790",
    "ss_seed": "3512259484",
    "ss_max_token_length": "225"

generating game assets like islands and objects in low polygonal landscapes, pixelated, 3D, and isometric styles.

Specifications:

• LoRA Type: LyCORIS/LoKr, Prodigy

• Trained Resolution: 1024 x 1024 pixels

• Tested Resolution: Up to 1600 pixels

• Training Data: 3 resumes each on 3000 images, totaling 9,000 images

Refiner example