Krea trained K2 image model from scratch on thousands of GPUs, now open source
“the whole idea was to like kind of bridge the gap between like LLM research and diffusion transformers”
image-generation
8 tracked signals on image-generation.
Krea open-sourced Krea 2 medium, arguing big labs sacrifice image diversity for consistency via mode collapse
“the easiest and most reliable way to render a person is render the most boring average person that exists”
Apple's revamped ImagePlayground framework lets apps generate photorealistic images via Private Cloud Compute.
“Your data is never stored or shared, even with Apple.”
AWS vLLM-Omni DLC enables unified text-to-image-to-video pipelines on SageMaker AI
Qualcomm VP argues image generation needs architectural innovation, not just scale
AWS shows how to run ComfyUI generative-AI workflows on SageMaker processing jobs for batch image generation at scale.
“ComfyUI is a node-based, visual workflow builder for generative AI that makes it straightforward to compose, test, and iterate on complex image, audio, or video pipelines without coding every step.”
Apple is deprecating the ImageCreator class, removing programmatic on-device image generation in OS version 27.
“the ImageCreator class is being discontinued and will no longer work in iOS 27, iPadOS 27, macOS 27, and visionOS 27 or later”
Systematic study finds no evidence AI labs train models to draw pelicans on bicycles better
“Pelicans aren't drawn any better than other animals. Bicycles aren't drawn any better than other vehicles. And no lab draws the combination better than its pelicans and bicycles already predict.”