This repository contains colab notebooks of interesting experiments on CLIP from different researchers and artists
After OpenAI released CLIP with its inference code and pre-trained model, the Twitter community is coming up with much amazing work on top of this model. Below are some of colab experiments for my personal collection.
Rayn Murdock came up with amazing idea of generating image from text prompt by guiding SIREN Network with CLIP.
Prompt: 'life during the plague'
| Notebook | Author |
|---|---|
Original notebook  |
Ryan Murdock |
| Simplified notebook |
luciddrains |
Another amazing idea by Rayn Murdock (to the best of my knowledge) where CLIP is used to optimize latent vetor of BigGAN in such a way that BigGAN generator output image which has higher similar with input prompt.
Prompt: 'A cloud in shape of donut'
Prompt: 'A colorful cartoon of a dog'
Credits: enricoros
| Notebook | Author |
|---|---|
| Original notebook |
Ryan Murdock |
| Simplified notebook |
eyalgruss.com |
Update: Wooh! Someone put BigSleep on web at https://dank.xyz/#
Generating an image from stylegan generator based on text prompt. Here similarity between generated image and input text is calculated using CLIP and it is combined with stylegan discriminator score to optimize input latent vector.
Prompt: 'A girl wearing crown'
Stylegan2 generator trained with FFHQ
Prompt: 'A queen sitting on chair'
Custom Stylegan2 generator used in thisanimedoesnotexist.ai
| Notebook | Author |
|---|---|
| Stylegan2/ada |
Levin Dabhi |
| Stylegan2 TADNE |
nagolinc |
You can search images from your image dataset or gallery by inputting text prompt
| Notebook | Author |
|---|---|
| Search from images pile |
Levin Dabhi |
| Search from Unsplash |
Vladimir Haltakov |