Checkpoint #9 – The week in AI art: Pixels, composable images, full humans and smooth animation
News & articles
The text-to-image revolution, explained – Excellent videos from Vox explaining the whole AI art thing. So good to see such excellent artists featured so prominently.
Text2Human – Text-to-image generator for full human image generation. Given a pose, it's able to generate diverse, high quality images of people with a wide range of clothing and styles. Huggingface Space | Paper.
Composable Diffusion – Diffusion model that understands spatial relationships between elements in the prompt. For example 'A red car parked in a desert and hills behind the car and aurora in the sky'. Colab | Paper.
Disco Diffusion 5.2 WarpFusion – Variant of Disco Diffusion from @devdef which uses optical flow to create much smoother animations.
JAX Diffusion 2.1 – Fantastic update from @huemin_art that enables much larger than normal images to be created by automatically stitching together smaller ones.
Featured notebook – Pixel Art Diffusion
Pixel Art Diffusion by @KaliYuga_ai is a flavour of Disco Diffusion with a custom model trained on pixel art landscapes. The result is an output with a fun pixel art aesthetic. The results aren't true pixel images, but it's a great starting point and can always be further processed using something like http://pixelatorapp.com.
Thanks for reading!
If you have anything you’d like to be featured or want to get in touch, give me a shout on Twitter or via email. Please also consider supporting me on Patreon so that I can spend more time creating content like this.