A collection of all the open-source AI tools I can wrest from the jaws of possible internet oblivion. Based on my post found here:
Table of Contents
- The Big Sleep
- VQGAN + Clip 2a. VQGAN+CLIP with Video Features
- Clip-Guided Diffusion
- Styledream Notebook
- Pixray PixelDraw
- Real-ESRGAN Practical Image Restoration
- ruDALLE
- Looking Glass AI
- JAX Clip-Guided Diffusion
- CLIP Guided Deep Image Prior
- Disco Diffusion
- Disco Diffusion Turbo
Description: The original Clip-guided AI Colab tool; uses Clip to guide BigGAN. Fairly outdated, but absolutely foundational to the entire art field. Author: Ryan Murdock (advadnoun on Twitter) Notebook Link
Gordian Gallery by KaliYuga, 2021
Description:
By far the most widely-used of the Clip-guided methods[Edit April 2022: very few people still seem to use this tool, but it's a foundational and very important one]; uses Clip to guide VQGAN. Author: Katherine Crowson (RiversHaveWings on twitter) Notebook Link (The notebook is in Spanish, but is easily translated to English using Chrome).
Description: What it says on the tin! VQGAN + Clip with added video features and also some really good demystifications of parameters. Author: @jotakrevs Notebook Link
Psychic Sermon in the Temple of the Mind by KaliYuga, 2021
Description: Produces absolutely unbelievable results, but very slow and resource-heavy. Uses Clip to guide a Diffusion model. Author: Katherine Crowson (RiversHaveWings on twitter) Notebook Link
Portraits of DC heroes in the style of Alex Ross *by .hack//HAMBRE
Description: Portraits only! Clip-guided StyleGAN model. Author: Ariel Ekgren (ArYoMo on twitter) Notebook Link
A synthwave great white shark by Viktor Alm
Description: A fork of NerdyRodent's fork of RiversHaveWings'VQGAN + Clip Notebook; uses VQGAN + Clip to produce pretty impressive pixel art. Author:Dribnet Notebook Link Extra! You can also start with an initial image in a different version of Pixray; if you use an image originally created by pixray, the results can be pretty trippy! For example:
Via the Colab Notebook's demo params
Other bits and pieces of this project can be found here.
Examples via the Colab Notebook
Description: A demo of Real-ESRGAN,which repairs many types of image degradation and also upscales the image 4x. Authors: Xintao Wang, Liangbin Xie, Chao Dong, Ying Shan Notebook Link
Description: A Fully-Trained, Accessible, Multi-Billion-Parameter Answer to OpenAI's DALL-E Model (in Russian) Authors: Sber AI and Sber Devices Notebook Link Bonus: A faster Colab notebook for free users
Description: An Accessible One-Shot Finetuner For RuDalle Authors: AI_Curio and Others Notebook Link
https://files.peakd.com/file/peakd-hive/kaliyuga/Ep3eMa4ioSQuG7mfrXBzwPsDe9wuNCG9sznjSWSE9eVHNXr8duCE4EN3fxbHZ8zAgdx.png By KaliYuga, February 2022
Description: [Via Gituhub] "The models are denoising diffusion probabilistic models (https://arxiv.org/abs/2006.11239), which are trained to reverse a gradual noising process, allowing the models to generate samples from the learned data distributions starting from random noise." Authors: Katherine Crowson and John David Pressman Notebook Link
Description: Creates soft, print-like images quite unlike what you generally get out of other CLIP-guided diffusion models Authors: Daniel Russell, based on work by Katherine Crowson and Ryan Murdock Notebook Link
Description: My daily drive; Very high quality image results very quickly; comes fully loaded with 3D animation capabilities and the latest Clip models. Authors: Somnai and Adam Letts Notebook Link
https://twitter.com/KaliYuga_ai/status/1502057904693723137?s=20&t=MLWq2-K9-Rc7Bu5IN9wZWg
Description: A fork of Disco which speeds up and cleans up 3D animation significantly. I also use this all the time! Author: Chris Allen Notebook Link Bonus by Chris Allen Notebook: Wiggle Standalone Keyframe Generator; lets you generate keyframe instructions that give you fine control over Turbo's motion.