|
--- |
|
language: |
|
- en |
|
license: creativeml-openrail-m |
|
thumbnail: "https://s3.amazonaws.com/moonup/production/uploads/1667942059199-6305d083df993a789e61126d.jpeg" |
|
tags: |
|
- stable-diffusion |
|
- text-to-image |
|
--- |
|
## Model description |
|
<b>isoCities</b> 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 "<b>an illustration of an isometric city, digital art, highly detailed</b>" in your prompts. |
|
|
|
[CKPT download link](https://huggingface.co/Astroboy/isoCities/blob/main/isometric_city%2BSD1.5.ckpt) |
|
|
|
## 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 |
|
 |
|
 |
|
 |
|
 |
|
 |
|
|
