one year on
xAI launches Grok-2 with image generation, limited guardrails
The new models, available in beta on X, offer frontier-class reasoning and FLUX.1-powered image creation that appears to lack guardrails around political figures.
xAI today released Grok-2 and Grok-2 mini in beta on X, marking a significant upgrade from Grok-1.5 with what the company calls ‘frontier capabilities in chat, coding, and reasoning.’ The new models are available to Premium and Premium+ subscribers on X, and xAI plans to offer them via its enterprise API later this month. An early version of Grok-2 had been tested on the LMSYS leaderboard under the alias ‘sus-column-r.’
The standout feature is image generation; text below sample prompts suggests Grok is using Black Forest Labs’ FLUX.1 model. Soon after release, X users began posting generated images of political figures, apparently with no guardrails around creating images of political figures. App researcher Nima Owji noted improved coding, writing, and news capabilities, though independent verification is sparse.
The lack of content restrictions raises immediate concerns about misinformation, especially with the U.S. presidential election approaching. It is unclear whether Grok-generated images embed metadata indicating AI provenance. TechCrunch said it has asked X how it plans to limit image generation for harmful purposes and will update the story if it hears back. The company indicated it will soon integrate Grok-2 into X features like search, post analytics, and replies, and will release a multimodal understanding preview.
Demonstrated Grok-2 generating political illustrations and images of real people, contrasting with ChatGPT's refusals.
Claimed Grok-2 shows improved coding, writing, and news generation capabilities, and will use FLUX.1 for images.
One year later — open only if you can handle spoilers
Grok-2's image generation soon faced widespread abuse, prompting xAI to eventually add some guardrails after public backlash. The incident became a key case study in the policy debate over AI-generated political content.