The Ai Artist Never Sleeps
There’s a new artist in town, and it never gets offended.
That’s because it was never programmed that way.
Using specialized neural networks from Megagloms like.Google. artists from all over the world are embarking on a new career; AI Artist. Using text inputs, some math and creative weighting of words and phrases to start, this new gen artist uses a Google Colab Notebook and with some effort and a few bucks, anyone can try their hand at this new form of computer generated imagery.
One of our first NFT purchases was from @Inner_Sanctum, an AI Artist who we found on Twitter. He was very receptive to us asking for a custom piece. and this is the result of us commissioning a woman with streams flowing around her.
Started the Journey 12-2021
We discovered this form of art of Twitter, where all the NFT action seems to be happening. Through links, talking to artists, buying some actual pieces, and listenign to a lot of Twitter Spaces, our creative curosity was piqued.
The first implementation was Big Sleep by Ryan Murdock/@advadnoun, which combined CLIP with an image generating GAN named BigGAN.
Katherine Crowson – Pioneer in AI Art plan to try and contact for interview.
Running a variant locally is possible, but needs serious setup and hardware. Prohibitive at this time, but keeping an eye on this space.
______
Here’s a Google Colab Notebook variant by “Nerdy Rodent.” Sheesh. 🙂
https://github.com/nerdyrodent/VQGAN-CLIP
DALL-E GITHUB REPO
https://github.com/openai/dall-e
MADE NOTE TO TRY THIS GOOGLE COLAB NOTEBOOK
https://colab.research.google.com/drive/1wkF67ThUz37T2_oPIuSwuO4e_-0vjaLs?usp=sharing
ANOTHER TRICK
One trick the community found to improve generated image quality is to simply add phrases that tell the AI to make a good image, such as artstationHQ or trending on /r/art. Trying that here:
<The Curator Notes: Using Unreal Engine, Fortnite, specific art stykes and more will get you very different looks.>
FROM – https://minimaxir.com/2021/08/vqgan-clip/ GOOD INFO
In this case, it’s unclear if the artstationHQ part of the prompt gets higher priority than the Salvador Dali part. Another trick that VQGAN + CLIP can do is take multiple input text prompts, which can add more control. Additionally, you can assign weights to these different prompts. So if we did cyberpunk forest by Salvador Dali:3 | artstationHQ, the model will try three times as hard to ensure that the prompt follows a Dali painting than artstationHQ.
CROWSON GUIDED DIFFUSION COLAB NOTEBOOK
https://colab.research.google.com/drive/1QBsaDAZv8np29FPbvjffbE1eytoJcsgA
PIXEL ART GENERATING COLAB NOTEBOOK.
https://colab.research.google.com/github/dribnet/clipit/blob/master/demos/PixelDrawer.ipynb
Where We Started – Dec. 2021 Notes
We discovered this form of art on Twitter (Now X), where all the NFT action seems to be happening. Through links, talking to artists, buying some actual pieces, and listenign to a lot of Twitter Spaces, our creative curosity was piqued.
The first implementation was Big Sleep by Ryan Murdock/@advadnoun, which combined CLIP with an image generating GAN named BigGAN.
Katherine Crowson – Pioneer in AI Art plan to try and contact for interview.
Running a variant locally is possible, but needs serious setup and hardware. Prohibitive at this time, but keeping an eye on this space.
______
Here’s a Google Colab Notebook variant by “Nerdy Rodent.” Sheesh. 🙂
https://github.com/nerdyrodent/VQGAN-CLIP
DALL-E GITHUB REPO
https://github.com/openai/dall-e
MADE NOTE TO TRY THIS GOOGLE COLAB NOTEBOOK
https://colab.research.google.com/drive/1wkF67ThUz37T2_oPIuSwuO4e_-0vjaLs?usp=sharing
ANOTHER TRICK
One trick the community found to improve generated image quality is to simply add phrases that tell the AI to make a good image, such as artstationHQ or trending on /r/art. Trying that here:
<The Curator Notes: Using Unreal Engine, Fortnite, specific art stykes and more will get you very different looks.>
FROM – https://minimaxir.com/2021/08/vqgan-clip/ GOOD INFO
In this case, it’s unclear if the artstationHQ part of the prompt gets higher priority than the Salvador Dali part. Another trick that VQGAN + CLIP can do is take multiple input text prompts, which can add more control. Additionally, you can assign weights to these different prompts. So if we did cyberpunk forest by Salvador Dali:3 | artstationHQ, the model will try three times as hard to ensure that the prompt follows a Dali painting than artstationHQ.
CROWSON GUIDED DIFFUSION COLAB NOTEBOOK
https://colab.research.google.com/drive/1QBsaDAZv8np29FPbvjffbE1eytoJcsgA
PIXEL ART GENERATING COLAB NOTEBOOK.Â
https://colab.research.google.com/github/dribnet/clipit/blob/master/demos/PixelDrawer.ipynb
0 Comments