Run your own DALL-E-like image generator

In this post, we'll look at getting setup with running your own A.I. image generator. For the July released retrieval-augmented diffusion models, you need scann in order to index the openimages dataset, which is only available on Linux. There are 2 image generation techniques possible with Latent Diffusion. I believe the txt2-img model that we'll setup first is what we are used to with other image generation tools online - it makes a super low res image clip thinks is a good prompt match and denoises and upscales it. The RAD model uses a configurable database of images as a reference AND does the diffusion like we are used to. Note, this model, especially when using the openimages training is best at recreating real things and doesn't seem very good and creating the weird images we're used to from the diffusion models. RDM. And there you have it - no more credits or monthly limits, just pure AI image generation all your own.