Skip to content

artistic stylegan3 datasets and models we created during our ongoing stylegan3 trip.

License

Notifications You must be signed in to change notification settings

edstoica/lucid_stylegan3_datasets_models

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

68 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Datasets and pretrained Models for StyleGAN3 ...

Dear arfiticial friend, this is a collection of artistic datasets and models that we have put together during our ongoing stylegan3 trip at the lucid layers studios. You can use the model snapshots for instant fun, or you can work with the source datasets to train your own models.
Some models include multiple snapshots. these can give interesting variations.

tip: best viewed maximized

Updates

This Document will be updated frequently since a lot of models are still in training on higher resolutions. You find information on each update in the "releases" section. (you may "watch" this repo for getting notified on new models)


... based on Wombo Dream:

All source images in this category were generated with Wombo Dream. thanks to the cheesy api implementation of adri326 we could remotely generate thousands of images. these images were cropped to center and scaled to 1024x1024. datasets in resolutions of 256, 512 and 1024 were generated and are available for download.
Most datasets are tied to one single textpromt with some minor variations. But you may mix them together in new datasets to train multi domain models with style mixing.

Thanks to the wombo creators for their generous quota limits.
Since wombo is a free service, we like to share all batches that have been created.

1. Mechanical devices from the future

Dataset
Name Mechanical devices from the future
Method Wombo Dream via Wombot
Image count 2169
Dataset download 256, 512, 1024
Samples
Model: 256 px
Method stylegan3-t, Transfer learning from Landscape256
Resolution 256x256
29 kimg
Download .pkl
 network-snapshot-000029 pkl
05 kimg
Download .pkl
  network-snapshot-000005 pkl

2. Vivid Flowers

Dataset
Name Vivid Flowers
Method Wombo Dream via Wombot
Image count 793
Dataset download 256, 512, 1024
Samples
Model: 256 px
Method stylegan3-t, Transfer learning from Landscape256
Resolution 256x256
68 kimg
Download .pkl
network-snapshot-000067 pkl
12 kimg
Download .pkl
network-snapshot-000012 pkl

3. Alien with Sunglasses

Dataset
Name Alien Sunglasses
Method Wombo Dream via Wombot
Image count 1600
Dataset download 256, 512, 1024
Samples
Model: 256 px
Method stylegan3-t, Transfer learning from Landscape256
Resolution 256x256
38 kimg
Download .pkl
network-snapshot-000038 pkl

4. forest daemons

Dataset
Name forest daemons
Method Wombo Dream via Wombot
Image count 794
Dataset download 256, 512, 1024
Samples
Model: 256 px
Method stylegan3-t, Transfer learning from Landscape256
Resolution 256x256
18 kimg
Download .pkl
network-snapshot-000018 pkl
03 kimg
Download .pkl
network-snapshot-000003 pkl

5. third eye watching you

Dataset
Name third eye watching you
Method Wombo Dream via Wombot
Image count 1363
Dataset download 256, 512, 1024
Samples
Model: coming soon ...

coming soon ...

6. mars spaceport

Dataset
Name mars spaceport
Method Wombo Dream via Wombot
Image count 710
Dataset download 256, 512, 1024
Samples
Model: coming soon ...

coming soon ...

7. scifi city

Dataset
Name scifi city
Method Wombo Dream via Wombot
Image count 1245
Dataset download 256, 512, 1024
Samples
Model: 256 px
Method stylegan3-t, Transfer learning from Landscape256
Resolution 256x256
210 kimg
Download .pkl
 network-snapshot-000210 pkl
018 kimg
Download .pkl
 network-snapshot-000018 pkl
013 kimg
Download .pkl
 network-snapshot-000013 pkl
008 kimg
Download .pkl
 network-snapshot-000008 pkl

8. scifi spaceship

Dataset
Name scifi spaceship
Method Wombo Dream via Wombot
Image count 1108
Dataset download 256, 512, 1024
Samples
Model: 256
Method stylegan3-t, Transfer learning from Landscape256
Resolution 256x256
168 kimg
Download .pkl
network-snapshot-000162 pkl
128 kimg
Download .pkl
network-snapshot-000128 pkl
13 kimg
Download .pkl
network-snapshot-000013 pkl

9. yellow comic alien

Dataset
Name yellow comic alien
Method Wombo Dream via Wombot
Image count 3984
Dataset download 256, 512, 1024
Samples
Model 256x256
Method stylegan3-t, Transfer learning from Landscape256
Resolution 256x256
19 kimg
Download .pkl
network-snapshot-000019 pkl
Model 512x512
Method stylegan3-t, Transfer learning from affhq
Resolution 512x512
236 kimg
Download .pkl
network-snapshot-000236 pkl
004 kimg
Download .pkl
network-snapshot-000004 pkl

10. eternal planet earth

Dataset
Name eternal planet earth
Method Wombo Dream via Wombot
Image count 1323
Dataset download 256, 512, 1024
Samples
Model: coming soon ...

11. mechanical landscape madness

Dataset
Name mechanical landscape madness
Method Wombo Dream via Wombot
Image count 1269
Dataset download 256, 512, 1024
Samples
Model 256x256
Method stylegan3-t, Transfer learning from Landscape256
Resolution 256x256
6 kimg
Download .pkl
network-snapshot-000006 pkl
5 kimg
Download .pkl
network-snapshot-000005 pkl

12. two aliens speaking

Dataset
Name two_aliens_speaking
Method Wombo Dream via Wombot
Image count 1159
Dataset download 256, 512, 1024
Samples
Model: coming soon...

Usage

We recomend to install the official StyleGAN3 repo to your local machine. then use the "visualizer.py" to start the gui. The GUI is very comfortable to use and allows easy visual inspection of the models, in realtime. (RTX card recomended).
If you have troubles for installing on windows with anaconda, try this edited enviroment.yml file. this works with the current pytorch release (cuda11).
Alternatively you may use a colab notebook to generate images/videos from the model. Just copy the links from this page to your favorite colab notebook.
Here is a basic notebook, pre-configured for many of our models:
Open In Colab

Progress

  • prepare all datasets in resolutions 256, 512, 1024
  • create a colab notebook for model testing
  • train some datasets in 256
  • train some datasets in 512
  • train some datasets in 1024
  • create multi domain datasets and models with style mixing

Contribution

If you do continue training on a model, or train a dataset in a high resolution, it would be great to include that in this list.
(please send me a link to your .pkl file in the "issues" tab)
Also, if you made some images or videos you like to share - we would love to see your work! put everything in the issues...

License

You are welcomed to use all files for your own purposes. we do not claim any copyrights, but it would be cool if you include a link to this repo in your work. Thank you.
(Terms and Conditions of Wombo, Nvidia and other contributors have to be considered seperately when doing commercial projects.)

About

artistic stylegan3 datasets and models we created during our ongoing stylegan3 trip.

Resources

License

Stars

Watchers

Forks

Packages

No packages published