xAI integrates new image generator Aurora into Grok
Until now, Elon Musk's xAI image generator in Grok ran on the Flux.1 model from Germany. Now it is moving into its own model, Aurora.
Grok's ability to generate images will in future be based on Aurora, a new and independent image generator. Previously, the Flux model from Black Forest Labs was responsible for image generation in the chatbot. The current status of the partnership with the German company is unknown. Aurora seems to have been given as few guard rails as the entire Grok chatbot.
Elon Musk had Grok developed by his company xAI, which was founded in the summer of 2023. It is an AI chatbot that is available to paying users of X. Grok is characterized by having fewer rules and limits when generating texts and images than other AI services. Musk says these are "humorous interactions" when the image generator hands Pokémon Pikachu a machine gun.
Grok delivers photorealistic images and has multimodal capabilities
Nothing seems to have changed in this respect. What has changed is the provider and the AI model that implements the requests. xAI speaks of an update for Grok, with an "autoregressive mixture of experts network". Aurora has been trained with billions of images from the internet so that it has a "deep understanding of the world". A Mixture of Experts model consists of several specialized models, each of which acts as an expert and of which only the required experts respond to a query. This is intended to improve performance while maintaining the same computing effort. Autoregressive means that the model predicts tokens and therefore works differently to previous diffusion models, which use noise to create images.
Thanks to Aurora, Grok should be able to generate photorealistic images and be very accurate in the implementation of the prompts. The model also works multimodally, so you can also enter and edit your own images, for example. The first testers have already used Aurora to generate and publish images of people such as Musk himself or Sam Altman, CEO of OpenAI, which look very realistic.
The previous image generator in Grok was developed by Black Forest Labs. Its founders were previously at Stability AI, which offers Stable Diffusion, a diffusion-based image generator. Flux.1 from Black Forest Labs claims to have a hybrid architecture consisting of multimodal and parallel diffusion transformer blocks and is said to have already outperformed models such as Midjourney v6.0, Dall-E 3 and SD3-Ultra in common benchmarks. This raises the question of whether Aurora is now even better or why xAI has switched providers.
(emw)