NovelAI's new v4 image generator is blowing my mind. These pictures were done without Lora, controlnet, Photoshop, inpainting, or any human effort beyond simple text, done on remote servers in under five seconds each. We are truly that much closer to a one-click magic pretty picture generator.
NovelAI *does* have inpainting and a form of controlnet, so these could have been cleaned up, but what's really impressing me right now is how lazy they are. The new v4 supposedly allows for advanced interactions that define an object and actions that can be acted upon it, as well as specific placement of things, like Flux and the newer image generators can do. The model is a preview, so all features haven't been implemented yet.
Just a couple of years ago, AI could not handle the things we're seeing here, like different body types, wrinkles, slit pupils, or the chainmail tie, the concept of which destroyed Stable Diffusion.
(Note: If you like text generation, NovelAI is among the best and image gen and text gen come bundled in the same paid service.)
Two men in a greenhouse, cartoon style, quirky style, looking at each other, gazing adoringly
man, Aziraphale, Good Omens, tan suit, brown vest, white collared shirt, tartan bowtie, hands clasped, smiling
Man, Crowley, Good Omens, yellow eyes, slit pupils, sunglasses, red hair, short hair, black jacket, quilted jacket, silver chain necktie, holding a potted plant, smiling