johko's picture
Update README.md
47c6a9a
metadata
library_name: keras
license: apache-2.0
pipeline_tag: text-to-image
tags:
  - keras-dreambooth
  - scifi

Model description

This model is a fine-tuned Stable Diffusion modeled, using the Dreambooth technique. It was trained on 15 images of Marvin the Paranoid Android, scraped from the Internet, as instance images and 205 robot images generated by Stable Diffusion as class images. You can find the full set here: Marvin the Paranoid Android

It was created by johko for the keras-dreambooth sprint.

Training procedure

This model was trained using the keras implementation of dreambooth. You can find the notebook to train these models and how to attend this sprint here.

Example Outputs

Marvin as Anime character Marvin as Lowpoly drawing Marvin as a pillow

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

Model Plot

View Model Plot

Model Image