library_name: keras
license: apache-2.0
tags:
- keras
- stable-diffusion
- text-to-image
- keras-dreambooth
- nature
Model description
This is a Keras Dreambooth model fine-tuned to images of galaxy mergers taken with the Hubble Space Telescope. In particular, I selected mergers at intermediate stages (3-4 here), where the two bodies of the merging galaxies are still recognizeable and/or a bridge of material connecting the two galaxies is clearly visible. Credit for the training images goes to ESA/Hubble: images can be found on their website selecting the category of interacting galaxies.
Sample Outputs
A few ouput samples of the Dreambooth model below.
prompt: image of sks galaxies merging in space
prompt: painting of sks galaxies merging in van gogh style, 8k, high quality, trending on artstation
Cosmic Hugs: an artistic rendition of galaxy collisions
Intended uses & limitations
This model has similar intended uses & limitations as any Stable Diffusion and Dreambooth models (see e.g. this model).
In particular, this model should be used with a prompt containing sks galaxies merging.
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
| Hyperparameters | Value |
|---|---|
| inner_optimizer.class_name | Custom>RMSprop |
| inner_optimizer.config.name | RMSprop |
| inner_optimizer.config.weight_decay | None |
| inner_optimizer.config.clipnorm | None |
| inner_optimizer.config.global_clipnorm | None |
| inner_optimizer.config.clipvalue | None |
| inner_optimizer.config.use_ema | False |
| inner_optimizer.config.ema_momentum | 0.99 |
| inner_optimizer.config.ema_overwrite_frequency | 100 |
| inner_optimizer.config.jit_compile | True |
| inner_optimizer.config.is_legacy_optimizer | False |
| inner_optimizer.config.learning_rate | 0.0010000000474974513 |
| inner_optimizer.config.rho | 0.9 |
| inner_optimizer.config.momentum | 0.0 |
| inner_optimizer.config.epsilon | 1e-07 |
| inner_optimizer.config.centered | False |
| dynamic | True |
| initial_scale | 32768.0 |
| dynamic_growth_steps | 2000 |
| training_precision | mixed_float16 |