Meta has released Muse Image, an AI image generation model that ranks second to OpenAI's GPT Image 2 on human preference benchmarks. The model is immediately available across Meta's apps and now includes a feature allowing users to generate images of real people by @-mentioning their public Instagram accounts without consent—a capability likely to face GDPR and EU AI Act scrutiny when Muse Image launches in Europe.
Summaries like this, in your inbox every morning.
Sign up free →What happened
Meta launched Muse Image, its first image generation model from Superintelligence Labs, which works as an AI agent that iteratively refines outputs by calling tools like code generation and web search. On the Image Arena evaluation platform, Muse Image ranks second in human preference scores for text-to-image and for both single- and multi-image editing, behind only OpenAI's GPT Image 2.
Why it matters
The model is now available in Meta AI app, on meta.ai, in Instagram Stories in the US, and in WhatsApp—reaching Meta's largest user bases. However, a newly introduced feature allows users to @-mention public Instagram accounts in prompts so Meta AI pulls photos from those profiles to generate images of that person with no consent required. The feature is on by default and appears set to face regulatory scrutiny in Europe under GDPR and the EU AI Act's deepfake labeling rules, which take effect August 2, 2026.
What to watch
Meta's invisible watermark system, Content Seal, survives cropping and compression, but whether a machine-readable watermark alone satisfies the EU AI Act's requirement that AI-generated images resembling real people be labeled in a way recognizable to affected people remains an open question. Images already generated will not be deleted even if users opt out of the feature.
Meta's Muse Image arrives after the company restructured its AI lab under Chief AI Officer Alexandr Wang, backed by billions in new investment. The model's agent-based approach—where it iteratively calls tools like code generation, web search, and local image editing rather than mapping prompts directly to images—mirrors OpenAI's GPT Image 2 architecture. On the Image Arena benchmark, Muse Image trails only GPT Image 2 in text-to-image and editing tasks, and Muse Video ranks third for text-to-video, though Meta acknowledges weaknesses in audio-video sync and fast motion.
The immediate controversy centers on the @-mention feature, which grants Meta the ability to synthesize images of real people from publicly available Instagram photos without individual consent. Meta has implemented an opt-out mechanism and a machine-readable watermark system (Content Seal), but the regulatory environment—particularly in Europe—appears hostile to this approach. The EU AI Act's Article 50 transparency rules, effective August 2, 2026, require that AI-generated or manipulated image, audio, and video content resembling real people and qualifying as deepfakes be clearly labeled as artificially created in a way recognizable to affected persons. Whether a post-hoc, machine-readable watermark satisfies this requirement remains contested, and GDPR's stricter data protection regime may force Meta to adopt an opt-in model or geographic restrictions if the feature launches in the EU at all.
No comments yet. Be the first to share your thoughts!
Log in to join the discussion
Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started FreeFree · takes 30 seconds · unsubscribe anytime
1 minute a day. The AI essentials.
200+ sources · Email / LINE / Slack
