Some Jupyter Notebooks, most hosted on google colab, that I think are cool. These can all run in the free version of google colab (on a Tesla K80).
- Stable Diffusion
- Fast Stable Diffusion
- Very Highly Recommended
- 25% faster, lower GPU memory usage, can fix faces using GFPGAN, can upscale 4x using ESRGAN. And very easy.
- Can only run on Colab hosted runtimes, not a local runtime.
- Super Simple Stable Diffusion
- An easier version of Stable Diffusion; still not the easiest. Not recommended.
- Balanced between creativity, speed, and resource use.
- Fast Stable Diffusion
- Super Simple Disco Diffusion
- Not updated to the latest version of Disco Diffusion
- Centipede Diffusion
- Based on Disco Diffusion v5.2 and Latent Diffusion
- Super Simple Latent Diffusion
- Not very creative, but can create fairly realistic images very quickly.
- Pixray
- 2 notebook locations:
- CLIP Guided Diffusion
- GFPGAN Inference
- Upscales faces in low-quality photos.
Here's a guide to connecting a local Jupyter runtime to the Google Colab interface. Make sure to get into the directory you want before starting jupyter, as Colab just runs in the directory you're already in.
To do a Jupyter runtime on a remote machine using ssh
for port forwarding, you can do this:
ssh -L 8888:localhost:8888 [email protected]
Just replace user
with the username for the remote machine you're running Jupyter on, and remote.address
with the address of the remote machine. Then, for a normal Jupyter lab interface, you can run jupyter-lab
and go here, or you can follow the guide above to use a remote runtime with the Google Colab interface.