Meta Superintelligence Labs officially launched Muse Image on July 7, 2026 — the lab's first proprietary AI image generation model — alongside a preview of Muse Video, marking a strategic shift away from third-party image providers toward a fully in-house AI media stack.
Key Highlights
- Muse Image ranks second on Arena's text-to-image leaderboard with an Elo score of 1,280, trailing only OpenAI's GPT Image 2
- The model also holds the second position in both single-image and multi-image editing arenas
- Muse Video previewed at third place in Arena's text-to-video leaderboard
- Available immediately on Meta AI, Instagram Stories (US), and WhatsApp in select regions
- Every generated image carries an invisible Content Seal watermark that survives cropping, compression, and resizing
What Makes Muse Image Different
Muse Image goes well beyond conventional prompt-to-image generation. The model employs agentic tool use — it can execute code for precise outputs like QR codes and data plots, and conduct live web searches to ground generations in current factual information, significantly improving accuracy on knowledge-intensive prompts.
A standout technical innovation is emergent self-refinement: Muse Image autonomously reviews its own outputs, making targeted edits or regenerating images entirely before delivering results. According to Meta, this behavior was not explicitly programmed — it emerged during reinforcement learning training because self-refinement consistently produced better images and earned higher reward signals.
The model also supports multi-reference image composition, allowing users to blend multiple input photos into a single coherent output, and deep instruction-following for complex prompts specifying camera angles, lighting conditions, artistic styles, and spatial arrangements. A room redesign feature pulls real products from Facebook Marketplace, bridging creative generation with e-commerce.
From Third-Party to In-House
Until now, Meta relied on Midjourney and Black Forest Labs to power its image generation features across Instagram, WhatsApp, and Facebook. Muse Image replaces those external providers with a fully proprietary model developed by Meta Superintelligence Labs, the AI division led by Chief AI Officer Alexandr Wang.
Muse Image is the second major model release from the lab, following Muse Spark — a large language model previewed in April 2026 as a Llama-series successor. The two models are designed to work together: Muse Spark provides joint agentic planning capabilities when paired with Muse Image for complex generative tasks.
Content Seal and Safety
Every image generated by Muse Image includes Content Seal, an invisible watermark engineered to persist through cropping, compression, color adjustments, and resizing. Meta provides a public detection tool at meta.ai/identification to verify whether an image was AI-generated.
Safety filters are built in to prevent generation of content that violates Meta's terms of service, including protections against child sexual abuse material (CSAM).
Privacy Controversy
The launch has not been without controversy. Instagram users can tag public profiles to generate AI images featuring other people — without notifying the person depicted. Meta's policy states users will not be informed when their photos are used in AI-generated content.
Critics have drawn parallels to Meta's historical privacy controversies, including Cambridge Analytica and earlier facial recognition disputes. Users who want to opt out must do so manually in their account settings.
Availability and Pricing
Muse Image is free to use within usage limits across Meta AI, Instagram Stories, and WhatsApp in select regions. Users exceeding those limits require a paid subscription to Meta's monthly plan, launched in May 2026. Facebook and Messenger integration is expected in the coming weeks.
Advertisers will gain access to Muse Image-powered creative variations within weeks, with the model set to integrate into Meta's Advantage+ automated advertising platform — representing a direct commercial return on Meta's AI infrastructure investment.
What's Next
Muse Video is expected to launch for creators and Meta AI users in the near term, promising strong prompt adherence, visual fidelity, and temporal consistency — though Meta acknowledges current limitations in audio-video synchronization and fast-motion sequences.
Meta has also signaled plans to offer Muse models via API to external developers, opening new revenue streams to offset the company's substantial AI infrastructure spending.
Source: Meta AI Blog