maroka wrote: Mon Oct 24, 2022 5:20 pm If this is Funny, let's see how Crazy Nurse look like?

it's funny because of the amount of hair, and her hat

i'm sure that's how they look fr

prompt is :
muscular strong looking impressive epic hyper realistic scary decadent, rough mechanical robot,
and it outputs me that :

The angel one is pretty cool!

If you're into anime and manga, then use Waifu Diffusion.
It's free.

Waifu Diffusion Colab - https://colab.research.google.com/drive ... J9QEgQBs_3

You can also find the link in discord server.

Prompt Guide -


Discord - https://discord.com/invite/touhouai
-----------------------------------------------------------------------
On the 1st of December, we will publicly release the fourth iteration of Waifu Diffusion.
The training data that will be used will be comprised of 10 to 20 million anime-styled images of variable resolutions.

A lot of ground work has been done in improving the training code in order to help make training as efficient as possible in order to save compute costs, as well as providing a lot of flexibility for everyone who is looking into doing their own finetuning work.

We are going to be using the following improvements in the fourth iteration:
- The penultimate layer of CLIP will be used. (credits go to NAI team and Google AI Research)
- CLIP token limit will be extended by 3 times.
- A newer VAE finetuned on 250k anime-styled images will be used, which will enhance the model's ability at producing finer details (eyes, fingers, etc)
- Open release of the dataset used.

We have a full overview of the list of improvements here: https://gist.github.com/harubaru/313eec ... d01f79a7e7

For the dataset, we are also looking for volunteer contributors to help curate and rate images in the dataset to help build a proper aesthetic filter. If you are interested in helping us, please let any staff member know and we will provide you with access to the contributor channel!

I would also like to thank the NovelAI team for publishing their improvements on Stable Diffusion, a lot of improvements that will be used here would not have been possible without their hard work and research.

And.. Lastly, I would like to personally thank @everyone for being here! Thank you for being with us as we train these models and release them openly for everyone.

https://discord.com/events/930499730843 ... 7423529010