justinpinkney / awesome-pretrained-stylegan2

A collection of pre-trained StyleGAN 2 models to download

Awesome Pretrained StyleGAN2

A collection of pre-trained StyleGAN2 models trained on different datasets at different resolution.

Note the readme is a bit out of date, there are more models linked in the issues.

• For StyleGAN3 models see: https://github.com/justinpinkney/awesome-pretrained-stylegan3
• For StyleGAN1 models see: https://github.com/justinpinkney/awesome-pretrained-stylegan)

If you have a publically accessible model which you know of, or would like to share please see the contributing section. Hint: the simplest way to submit a model is to fill in this form.

Table of Contents

• Models

  • car (config-e)
  • car (config-f)
  • cat
  • church
  • faces (FFHQ config-e)
  • faces (FFHQ config-e 256x256)
  • faces (FFHQ config-f)
  • faces (FFHQ config-f 512x512)
  • horse
  • Imagenet
  • WikiArt
  • Anime portraits
  • microscope images
  • wildlife
  • modern art
  • trypophobia
  • Abstract art
  • Maps
  • cakes
  • CIFAR 10
  • CIFAR 100
  • faces (FFHQ slim 256x256)
  • obama
  • grumpy cat
  • panda
  • fursona
  • my little pony
  • painting faces
  • ukiyoe faces
  • beetles
  • textures
  • more abstract art
  • flowers
  • Doors
  • floor plans
  • figure drawings

• Notes

• Contributing

car (config-e)

• Style mixing example, interpolation video
• Dataset: LSUN Car
• Resolution: 512x512 config: e
• Author: NVIDIA
• Download link
• StyleGAN2 implementation: https://github.com/NVlabs/stylegan2
• Licence: Nvidia Source Code License-NC
• Source

car (config-f)

• Style mixing example, interpolation video
• Dataset: LSUN Car
• Resolution: 512x512 config: f
• Author: NVIDIA
• Download link
• StyleGAN2 implementation: https://github.com/NVlabs/stylegan2
• Licence: Nvidia Source Code License-NC
• Source

cat

• Style mixing example, interpolation video
• Dataset: LSUN Cat
• Resolution: 256x256 config: f
• Author: NVIDIA
• Download link
• StyleGAN2 implementation: https://github.com/NVlabs/stylegan2
• Licence: Nvidia Source Code License-NC
• Source

church

• Style mixing example, interpolation video
• Dataset: LSUN Church
• Resolution: 256x256 config: f
• Author: NVIDIA
• Download link
• StyleGAN2 implementation: https://github.com/NVlabs/stylegan2
• Licence: Nvidia Source Code License-NC
• Source

faces (FFHQ config-e)

• Style mixing example, interpolation video
• Dataset: FFHQ
• Resolution: 1024x1024 config: e
• Author: NVIDIA
• Download link
• StyleGAN2 implementation: https://github.com/NVlabs/stylegan2
• Licence: Nvidia Source Code License-NC
• Source

faces (FFHQ config-e 256x256)

• Style mixing example, interpolation video
• Dataset: FFHQ
• Resolution: 256x256 config: e
• Author: Justin Pinkney
• Download link
• StyleGAN2 implementation: https://github.com/justinpinkney/stylegan2
• Notes: Trained to FID 11.2 from scratch for 3810 kimg
• Licence: CC BY-NC-SA 4.0
• Source

faces (FFHQ config-f)

• Style mixing example, interpolation video
• Dataset: FFHQ
• Resolution: 1024x1024 config: f
• Author: NVIDIA
• Download link
• StyleGAN2 implementation: https://github.com/NVlabs/stylegan2
• Licence: Nvidia Source Code License-NC
• Source

faces (FFHQ config-f 512x512)

• Style mixing example, interpolation video
• Dataset: FFHQ
• Resolution: 512x512 config: f
• Author: aydao
• Download link
• StyleGAN2 implementation:
• Licence: Public Domain
• Source

horse

• Style mixing example, interpolation video
• Dataset: LSUN Horse
• Resolution: 256x256 config: f
• Author: NVIDIA
• Download link
• StyleGAN2 implementation: https://github.com/NVlabs/stylegan2
• Licence: Nvidia Source Code License-NC
• Source

Imagenet

• Style mixing example, interpolation video
• Dataset: Image net
• Resolution: 512x512 config: Unknown
• Author: Shawn Presser
• Download link
• StyleGAN2 implementation: Unknown
• Notes: Trained using TPUs
• Licence: Unknown
• Source

WikiArt

• Style mixing example, interpolation video
• Dataset: WikiArt
• Resolution: Unknown config: Unknown
• Author: Peter Baylies
• Download link
• StyleGAN2 implementation: https://github.com/pbaylies/stylegan2
• Notes: Conditional
• Licence: Unknown
• Source

Anime portraits

• Style mixing example, interpolation video
• Dataset: Danboru
• Resolution: 512x512 config: f
• Author: Aaron Gokaslan
• Download link
• StyleGAN2 implementation: Unknown
• Licence: Unknown
• Source

microscope images

• Style mixing example, interpolation video
• Dataset: Unknown
• Resolution: 512x512 config: Unknown
• Author: Michael Friesen
• Download link
• StyleGAN2 implementation: Unknown
• Licence: Unknown
• Source

wildlife

• Style mixing example, interpolation video
• Dataset: Unknown
• Resolution: Unknown config: Unknown
• Author: Michael Friesen
• Download link
• StyleGAN2 implementation: Unknown
• Licence: Unknown
• Source

modern art

• Style mixing example, interpolation video
• Dataset: Unknown
• Resolution: Unknown config: Unknown
• Author: Michael Friesen
• Download link
• StyleGAN2 implementation: Unknown
• Licence: Unknown
• Source

trypophobia

• Style mixing example, interpolation video
• Dataset: https://drive.google.com/file/d/1u_fLHmO6JuJlBTQIKRGgl4PeBKbBu9GJ/view
• Resolution: 1024x1024 config: f
• Author: Sid Black
• Download link
• StyleGAN2 implementation: https://github.com/NVlabs/stylegan2
• Licence: Nvidia Source Code License-NC
• Source

Abstract art

• Style mixing example, interpolation video
• Dataset: Frea Buckler artwork
• Resolution: 1024x1024 config: f
• Author: Derrick Schultz
• Download link
• StyleGAN2 implementation: RunwayML
• Notes: Based on Frea Buckler’s artwork from her Instagram account (purposefully undertrained to be abstract and not infringe on the artist’s own work)
• Licence: Unknown
• Source

Maps

• Style mixing example, interpolation video
• Dataset: Maps
• Resolution: 1024x1024 config: f
• Author: Topi Tjukanov
• Download link
• StyleGAN2 implementation: https://github.com/NVlabs/stylegan2
• Licence: Unknown
• Source

cakes

• Style mixing example, interpolation video
• Dataset: Internet scraped cakes
• Resolution: 256x256 config: e
• Author: Justin Pinkney
• Download link
• StyleGAN2 implementation: https://github.com/justinpinkney/stylegan2
• Notes: Trained from scratch to FID 13.6
• Licence: CC BY-NC-SA 4.0
• Source

CIFAR 10

• Style mixing example, interpolation video
• Dataset: CIFAR 10
• Resolution: 32x32 config: see paper
• Author: mit-han-lab
• Download link
• StyleGAN2 implementation: https://github.com/mit-han-lab/data-efficient-gans/tree/master/DiffAugment-stylegan2
• Notes: Trained with DiffAugment, FID (10k) = 9.89
• Licence: Unknown
• Source

CIFAR 100

• Style mixing example, interpolation video
• Dataset: CIFAR 100
• Resolution: 32x32 config: see paper
• Author: mit-han-lab
• Download link
• StyleGAN2 implementation: https://github.com/mit-han-lab/data-efficient-gans/tree/master/DiffAugment-stylegan2
• Notes: Trained with DiffAugment, FID (10k) = 15.22
• Licence: Unknown
• Source

faces (FFHQ slim 256x256)

• Style mixing example, interpolation video
• Dataset: FFHQ
• Resolution: 256x256 config: slim
• Author: mit-han-lab
• Download link
• StyleGAN2 implementation: https://github.com/mit-han-lab/data-efficient-gans/tree/master/DiffAugment-stylegan2
• Notes: Trained with DiffAugment, FID = 3.81
• Licence: Unknown
• Source

obama

• Style mixing example, interpolation video
• Dataset: 100 images of Barack Obama
• Resolution: 256x256 config: f
• Author: mit-han-lab
• Download link
• StyleGAN2 implementation: https://github.com/mit-han-lab/data-efficient-gans/tree/master/DiffAugment-stylegan2
• Notes: Trained with DiffAugment, FID (5k) = 46.87
• Licence: Unknown
• Source

grumpy cat

• Style mixing example, interpolation video
• Dataset: 100 images of Grumpy Cats
• Resolution: 256x256 config: f
• Author: mit-han-lab
• Download link
• StyleGAN2 implementation: https://github.com/mit-han-lab/data-efficient-gans/tree/master/DiffAugment-stylegan2
• Notes: Trained with DiffAugment, FID (5k) = 27.08
• Licence: Unknown
• Source

panda

• Style mixing example, interpolation video
• Dataset: 100 images of pandas
• Resolution: 256x256 config: f
• Author: mit-han-lab
• Download link
• StyleGAN2 implementation: https://github.com/mit-han-lab/data-efficient-gans/tree/master/DiffAugment-stylegan2
• Notes: Trained with DiffAugment, FID (5k) = 12.06
• Licence: Unknown
• Source

fursona

• Style mixing example, interpolation video
• Dataset: ~55k SFW images from e621.net
• Resolution: 512x512 config: TBC
• Author: arfa
• Download link
• StyleGAN2 implementation: Unknown
• Notes: Trained using TPUs
• Licence: Unknown
• Source

my little pony

• Style mixing example, interpolation video
• Dataset: ~104k SFW images from Derpibooru
• Resolution: 1024x1024 config: TBC
• Author: arfa
• Download link
• StyleGAN2 implementation: Unknown
• Notes: Trained using TPUs
• Licence: Unknown
• Source

painting faces

• Style mixing example, interpolation video
• Dataset: MetFaces
• Resolution: 1024x1024 config: f
• Author: AK
• Download link
• StyleGAN2 implementation: Unknown
• Licence: Unknown
• Source

ukiyoe faces

• Style mixing example, interpolation video
• Dataset: 5000 faces aligned and detected from ukiyoe images
• Resolution: 256x256 config: slim
• Author: Justin Pinkney
• Download link
• StyleGAN2 implementation: https://github.com/justinpinkney/data-efficient-gans/
• Notes: Fine tuned from ffhq-256-slim, used DiffAugment for training, FID = 12.74
• Licence: CC BY-NC-SA 4.0
• Source

beetles

• Style mixing example, interpolation video
• Dataset: Biologia Centrali-Americana :zoology, botany and archaeology
• Resolution: 1024x1024 config: f
• Author: Bernat Cuni
• Download link
• StyleGAN2 implementation: Unknown
• Licence: Unknown
• Source

textures

• Style mixing example, interpolation video
• Dataset: Describable Textures Dataset (DTD)
• Resolution: 1024x1024 config: f
• Author: Bernat Cuni
• Download link
• StyleGAN2 implementation: Unknown
• Licence: Unknown
• Source

more abstract art

• Style mixing example, interpolation video
• Dataset: 14,305 abstract paintings
• Resolution: 512x512 config: f
• Author: Nick Saraev
• Download link
• StyleGAN2 implementation: Unknown
• Notes: Fine tuned from FFHQ 512 model
• Licence: Unknown
• Source

flowers

• Style mixing example, interpolation video
• Dataset: Oxford flowers 102 prepped with u^2net
• Resolution: 256x256 config: slim
• Author: Justin Pinkney
• Download link
• StyleGAN2 implementation: https://github.com/justinpinkney/data-efficient-gans/
• Notes: Fine tuned from ffhq-256-slim, used DiffAugment for training, FID = 12.20
• Licence: CC BY-NC-SA 4.0
• Source

Doors

• Style mixing example, interpolation video
• Dataset: 5k architectural elements from Barcelona
• Resolution: 256x256 config: f
• Author: Vasily Korf
• Download link
• StyleGAN2 implementation: https://github.com/NVlabs/stylegan2
• Notes: styleGAN trained on architectural elements to create Art Nouveau doors
• Licence: Unknown
• Source

floor plans

• Style mixing example, interpolation video
• Dataset: Floor Plans (Instagram scrapped)
• Resolution: 1024x1024 config: a
• Author: Mayur Mistry
• Download link
• StyleGAN2 implementation: https://github.com/NVlabs/stylegan
• Licence: Unknown
• Source

figure drawings

• Style mixing example, interpolation video
• Dataset: Previous models made from the drawings from my blog
• Resolution: 1024x1024 config: f
• Author: Krrrl
• Download link
• StyleGAN2 implementation: Runway ML
• Licence: Unknown
• Source

Notes

• The configuration "slim" refers to the reduced feature map model used in the Karras limited data and Zhao data efficient papers.
• Each row in the sample grids above use a different level of trunction: 0.25, 0.5, 0.75, 1 from top to bottom.
• Style mixing figure and interpolation video generated using truncation of 0.75

Contributing

TLDR: You can either edit the models.json file or fill out this form.

This readme is automatically generated using Jinja, please do not try and edit it directly. Information about the models is stored in models.json please add your model to this file. Preview images are generated automatically and the process is used to test the link so please only edit the json file.

awesome-pretrained-stylegan2's Issues

Impressionist landscape painting

!gdown --id 1AHKufJ5o4dQFThFN0A-pJRRC2lzxEH2J

Source: https://towardsdatascience.com/ganscapes-using-ai-to-create-new-impressionist-paintings-d6af1cf94c56

Add corgi model

https://github.com/seawee1/Did-Somebody-Say-Corgi

Add fine tuned from field

Add details of if the network was trained using transfer learning, and if so what network it was from.

This will be useful for using the models for fun like interpolation and blending (e.g. https://www.justinpinkney.com/stylegan-network-blending/)

This will also be important when adding models in #7

Add links to Doron Adler's fine tuned models

Doron Adler shared a link to a bunch of fine tuned models i.e. transfer learning from a base model for just a few ticks.

https://twitter.com/Norod78/status/1255237713118343168

These often give weird but interesting results, doesn't quite feel right to put them in the main list, but there should be a link somewhere.

[Noob question] What is the difference between config-f and config-e ?

StyleGAN Human

https://github.com/stylegan-human/StyleGAN-Human

256x256 FFHQ StyleGAN-V2

Hi,
I was wondering if you're aware of a pretrained 256x256 FFHQ StyleGANv2 model available? I think it would be really be helpful for tasks where the full model is too big to start with.

What's the implementation of ImageNet pretrained model?

Add interpolation videos

Generate some interpolation videos for the different models

https://github.com/streamlit/demo-face-gan

this is quite good - I wonder how hard it would be to fire up this with swapped models.

Beach model

https://thisbeachdoesnotexist.com/

Getting permissions to use in professional work

Hi,

Firstly thanks for the repo, it is indeed awesome :)

A couple of questions:

• do you have a contact for @Krrrl, author of the lifedrawings dataset? I'm struggling to reach them.

• do you have any thoughts on ethical issues around using the lifedrawings dataset in 'professional' work? I'm a musician working on an audiovisual commission (aimed much more at a musical audience than a visual art audience) and I REALLY want to use that dataset for one sub-piece in it. I would be significantly altering the appearance of the images, but still using them without the author's permission feels a bit wrong. I was wondering if you knew of their intention in making the dataset publicly available, if there was any intention behind it all.

Best wishes,

Mark

Add models from Stylegan2 ADA

How to run the script?

Can anyone tell me how to run the script to generate a new image? I've already downloaded the weights. How to create new images using GAN with the script?

stylegan3

Dear Justin,

Thank you for creating the great repo. Do you have plan to support stylegan3?

Thank you for your help.

Best Wishes,

Alex

Several StyleGAN2 models to add

https://github.com/jeffheaton?tab=repositories&q=pretrained&type=&language=
Minecraft 1024px, Fish 256px, 1970s Sci-Fi 1024px, Christmas 256px

Judging by the fakes000000.png images from each, they weren't using transfer learning.

https://github.com/joshchen984/pretrained-stylegan2-hedgehogs
Hedgehogs 256px

https://github.com/dobrosketchkun/NeuralKuvshinov
https://drive.google.com/drive/folders/1T6BNlyPpvLsxXI-gkCepR7S_nFcSi63o (larger number = trained longer)
kuvshinov_ilya instagram artwork, transfer trained from stylegan2-ffhq-config-f.pkl

Is there any pre-trained model for lsun bedroom and landscpae?

Hi, nice work. I wonder is there any pre-trained model for lsun bedroom and landscpae? Thanks a lot.

please explain how can i use these to make videos. im a newbie

Add architecture model

From Gabriel Syme on the TPU podcast discord:

Hello everyone! Given that I'm still struggling to get colab TPUs going, and I don't think I'll be spending more credits on vastai for this right now, I decided to share my Stylegan2-ada-pytorch model trained on about 3,000 real and stylistic images of architectural buildings, across 18 diverse architectural practices. Class (practice) information was not used during training, this is a 512x512 model and has been trained for a bit over 3kimg. I will eventually work on building a larger data set in the future across more practices but for now this is it.

The model isn't as good as I would like it to be, but it can still generate quite nice images. Maybe someone finds a cool seed!

pkl file: https://storage.googleapis.com/pretrained-stylegan2/stylegan-ada-pytorch-buildings-3kimg.pkl
dataset: https://storage.googleapis.com/pretrained-stylegan2/Buildings.rar

This is a model from the new stylegan-ada-pytorch repo

About blending 'ffhq' model and 'anime face' model

Hi ,Really thanks for your work.
From your blog I notice that blending needs a base model and a fine-tuned model from that. So I try it and, I want to know that if there is a fine-tuned model for Anime Face in stylegan2 ?
The Aaron Gokaslan's anime mode is not fine-tuned and can't be used for blending with ffhq. So if it's possible to use a 512-f-ffhq model fine-tuning an anime-face dataset so that the new anime face model can be used for blending?

Clothing (lookbook) model

https://drive.google.com/uc?export=download&id=1-F-RMkbHUv_S_k-_olh43mu5rDUMGYKe

via: https://github.com/mfrashad/ClothingGAN

gen-art-project project

new edition from @stuartwaller

https://github.com/StuartWaller/gen-art-project

model
https://drive.google.com/file/d/1AUVmc1wYSJ01JFvqGa4Bb0OERLdA0fXg/view?usp=sharing

generated images using the provided pretrained model contain colorful specks.???

the generated images using the provided pretrained model in these links contain colorful specks. For example, dataset: faces (FFHQ config-f). Anybody knows what's wrong with this?

Cannot process pkl files using legacy.load_network_pkl

I am running a notebook called "structured dreaming". The notebook loads the pkl file using this:

print('Loading networks from "%s"...' % network_pkl) device = torch.device('cuda') with dnnlib.util.open_url(network_pkl) as f: G = legacy.load_network_pkl(f)['G_ema'].to(device) # type: ignore

For network_pkl, this works:

https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada-pytorch/pretrained/ffhq.pkl'

but when I try one of your pkl files, for instance:

https://github.com/justinpinkney/awesome-pretrained-stylegan2#faces-FFHQ-config-f-512x512 (after downloading it first and then putting it into Gdrive), I get:

`Loading networks from "/content/drive/MyDrive/pretrained/ffhq-512-avg-tpurun1.pkl"...

---

ValueError Traceback (most recent call last)

in ()
5 device = torch.device('cuda')
6 with dnnlib.util.open_url(network_pkl) as f:
----> 7 G = legacy.load_network_pkl(f)['G_ema'].to(device) # type: ignore
8 #
9 # Read the data from the file

1 frames

/content/stylegan2-ada-pytorch/legacy.py in convert_tf_generator(tf_G)
153 unknown_kwargs = list(set(tf_kwargs.keys()) - known_kwargs)
154 if len(unknown_kwargs) > 0:
--> 155 raise ValueError('Unknown TensorFlow kwarg', unknown_kwargs[0])
156
157 # Collect params.

ValueError: ('Unknown TensorFlow kwarg', 'resolution_h')

`

Do you have any suggestions? Your list of pretrained networks is great!

I resume training and init fakes has wrong colours

Hi,
i want to resume training with car config-f, when I started the training after the first tick, if i preview the init fakes they have messed up colours. Is this normal or is there some conflict with a colour space I can choose somehow?
Thanks