I released the dragon and pointed @arrendek towards VQGN+CLIP deep learning image generation. He has been experimenting on qt3 user names. It’s like Frankenstein, but with 2D art.
If you want to try this yourself, here’s what you need to do:
- Install the NVidia CUDA toolkit: CUDA Toolkit 11.5 Downloads | NVIDIA Developer
- Install Anaconda Python: Anaconda | Individual Edition
- Install Git: https://git-scm.com/
- Follow these instructions: GitHub - nerdyrodent/VQGAN-CLIP: Just playing with getting VQGAN+CLIP running locally, rather than having to use colab.
- This requires a lot of VRAM. With 10Gb, you’ll max out at 448x448 images. To mitigate that, you can use ESRGAN, to use deep learning to upscale the images: GitHub - xinntao/Real-ESRGAN: Real-ESRGAN aims at developing Practical Algorithms for General Image Restoration.
Examples (note it uses a random seed and is different with each running):
python generate.py -p "An impressionist painting of a burning cat eating apples on a space station | trending on Artstation" -s 384 384 -o "cat1.png"
python generate.py -p "Tom playing an extreme sports game and streaming for an online audience|character design on Behance HD" -s 448 448 -o "tc2.png"
Why “trending on Artstation” and “character design on Behance HD”? Best to check out these articles:
Dave examining a packet of Oreos for @Dave_Perkins