Model description
isoCities v1
This model trained based on Stable Diffusion 1.5 model to create isometric cities, venues, etc more precisely. Trained isometric city model merged with SD 1.5 with Automatic1111's checkpoint merger tool (Couldn't remember exactly the merging ratio and the interpolation method)
Use "an illustration of an isometric city, digital art, highly detailed" in your prompts.
Intended uses & limitations
Do not use for commercial purposes.
Training procedure
Trained on a WSL2 Ubuntu machine with DreamBooth.
Training hyperparameters
The following hyperparameters were used during training:
--with_prior_preservation --prior_loss_weight=1.0
--instance_prompt="an illustration of an isometric city, digital art, highly detailed"
--class_prompt="isometric city"
--resolution=512
--train_batch_size=1
--mixed_precision="fp16"
--gradient_accumulation_steps=1 --gradient_checkpointing
--use_8bit_adam
--learning_rate=5e-6
--lr_scheduler="constant"
--lr_warmup_steps=0
--num_class_images=50
--max_train_steps=1500
Training results
Sample images
