stylegan2-ada VS clip-guided-diffusion

If you smudge just a few locations I doubt it would fool a simple discriminator. You could also train a discriminator that is robust to post-processing by using augmentations. This was popular with StyleGAN models: https://github.com/NVlabs/stylegan2-ada

https://github.com/NVlabs/stylegan2-ada + clip guided diffusion

StyleGAN2-ADA is a neural network good at learning styles from images, you can give it a dataset and 'learn' its style into a file (a trained model). In this example, I load a model and given a random seed, generate a random texture which is applied to the object's material.

You have to modify the code, it's line 60 in https://github.com/NVlabs/stylegan2-ada/blob/main/generate.py

no i dont.. but as for training - i just use default tf stylegan2-ada repo ( https://github.com/NVlabs/stylegan2-ada )

You should check stylegan-2 ada, it works on colab and can be trained less than 12 hours tensorflow implementation

According to his bio he uses "clip-guided diffusion". Never heard of it before, but it appears to not use GANs. Text model and image classifier.

But yeah this software generates similar but to be fair not nearly as “aesthetic” gifs with a single terminal command and actually 0 photoshop.

I used CLIP guided diffusion to generate the image (see OpenAi CLIP).

I used this https://github.com/afiaka87/clip-guided-diffusion. No reference at all only a prompt "Steampunk town"

I used this method. It consumes about 8gb of vram and takes about 20 minutes to generate 1 image. You can also import it in colab. And if you get an unlucky seed, you have to reset the timer and start crafting your items again.