
TL;DR
Meta's Muse Image is now in Meta AI, but it is not a public model API. Here is what the launch confirms, what remains preview-only, and how developers should evaluate it.
Last updated: July 9, 2026
Meta has introduced two related media models from Meta Superintelligence Labs: Muse Image and Muse Video. "Meta Muse" is not one product, SDK, or downloadable model.
The practical update is simpler. Muse Image is a consumer-facing image-generation capability available in Meta AI and selected Meta surfaces. Muse Video is an early preview, not a generally available developer product. Meta has also announced a public-preview Meta Model API for the separate Muse Spark 1.1 reasoning model. That does not confirm a Muse Image API.
Meta describes Muse Image as an agentic image system that can reason, use tools, refine work, compose multiple references, and use social context. Do not turn that announcement into imaginary endpoints, pricing, or deployment promises.
| Question | What Meta has confirmed as of July 9, 2026 |
|---|---|
| What is Muse Image? | Meta Superintelligence Labs' image-generation model, used in Meta AI and selected Meta products. |
| Can people use it today? | Yes, in the Meta AI app and on meta.ai, plus Instagram Stories in the US and WhatsApp in limited countries. |
| Is there a public Muse Image API? | No public API announcement for Muse Image was found in Meta's launch materials. |
| Are weights or a model card available? | Meta's launch materials do not confirm public weights or a public Muse Image model card. |
| What about Muse Video? | It is an early preview and is described as coming soon to creators and Meta AI. |
| What is available through the Meta Model API preview? | Muse Spark 1.1, Meta's multimodal reasoning model, according to Meta's July 9 announcement. |
The right mental model is: Muse Image is presently a product capability to evaluate in Meta's apps, while Muse Spark 1.1 is the model Meta has explicitly positioned for developers through its API preview.
Most image tools are presented as a prompt in and image out interaction. Meta describes a broader loop for Muse Image. Before and during image generation, it can plan, call tools, use code for exact visual elements, seek external context, and run self-refinement steps.
When an image request requires a correct plot, a scannable QR code, current factual context, or several visual references arranged precisely, the hard problem is coordinating several kinds of work and judging whether the output satisfies the request.
Meta's research announcement gives three concrete examples of this system behavior:
These are Meta product claims, not independently reproduced benchmarks. Still, the architecture is notable: visual AI quality is increasingly tied to the workflow around generation, including references, verification, layout reasoning, and selective tools.
Newsletter
Get the weekly deep dive
Tutorials on Claude Code, AI agents, and dev tools, delivered free every week.
From the archive
Jul 9, 2026 • 7 min read
Jul 9, 2026 • 5 min read
Jul 9, 2026 • 8 min read
Jul 8, 2026 • 7 min read
Meta says Muse Image can combine many input references, including people, objects, clothes, styles, and environments, with text and images interleaved in a prompt. Its consumer product announcement also describes using multiple photos and @-mentioning public Instagram accounts in Meta AI, subject to the account controls Meta links from the feature.
Multi-reference work is where generic image generation often becomes unreliable: identity drifts, a product changes shape, or a scene loses an important object.
For a developer evaluating consumer image systems, make the test set reflect that reality. Build a small set of repeatable cases:
Keep the originals, prompts, outputs, and a human pass-fail note. It measures whether the system holds onto what your workflow needs.
Meta says Muse Image is available in the Meta AI app and on meta.ai, with Instagram Stories availability in the United States and WhatsApp availability in limited countries. The company says Facebook and other surfaces are coming later. Meta also says everyday creation is free, with additional creation available through its subscription plans.
Those statements are useful for product exploration, not a promise that an external application can automate the experience or embed the model. There is no public Muse Image API syntax to copy from the launch posts, nor confirmed public weights, pricing units, rate limits, model card, or enterprise data terms.
That is the current boundary. Keep internal image-generation abstractions provider-neutral until Meta publishes a developer contract. A clean interface for a media job, references, settings, output asset, and review status is more durable than coding against an unannounced interface.
Meta's research post describes Muse Video as sharing a pretraining base with Muse Image and supporting native audio. It calls the release an early preview and notes active work on audio-video synchronization and physically accurate fast motion. Meta says Muse Video is coming soon to creators and Meta AI.
Muse Video is not a production dependency. Do not schedule a video pipeline around it, quote a public API surface, or promise a launch date based on a preview.
Meta says images generated in Meta AI and on meta.ai carry its invisible Content Seal watermark. The company says the signal is designed to survive cropping, compression, resizing, and screenshots, and it is previewing an identification tool for checking whether an image carries the watermark.
If your team publishes generated visuals, provenance should be included in the acceptance checklist alongside quality, rights, approvals, and accessibility. Content Seal is Meta's system, not a universal guarantee that any image can be attributed or every transformation detected.
Use a short evaluation loop:
For the announced model platform, read Meta's Muse Spark 1.1 and Meta Model API announcement. Its API preview is for Muse Spark 1.1, not Muse Image.
Evaluate the experience that exists today, but keep a hard line between a consumer launch and a supported developer platform. Muse Image may become an integration target later. On July 9, 2026, it is an image product to explore, not a public API to build against.
Not exactly. Meta has announced Muse Image, Muse Video, and Muse Spark. Muse Image and Muse Video are media-generation models, while Muse Spark 1.1 is a separate multimodal reasoning model. "Meta Muse" is informal shorthand and can be ambiguous.
Meta's July 2026 Muse Image launch materials do not announce a public Muse Image API. Meta's separate public-preview Meta Model API announcement is for Muse Spark 1.1.
No. Meta describes Muse Video as an early preview that is coming soon to creators and Meta AI. It has not announced general developer availability in the sources below.
Meta's launch materials do not confirm public weights or a public Muse Image model card. Do not assume that either exists until Meta publishes it.
Meta says Muse Image is available in the Meta AI app and on meta.ai, Instagram Stories in the US, and WhatsApp in limited countries, with more Meta surfaces planned.
Read next
The latest GPT Image 2 prompt-library repos are not just galleries. They point at a practical workflow for repeatable visual systems, agent-friendly templates, and cheaper creative iteration.
7 min readInception Labs launched Mercury, the first commercial-grade diffusion large language model. It generates over 1,000 tokens per second on standard Nvidia hardware by replacing autoregressive generation with a coarse-to-fine diffusion process.
7 min readA practical comparison of OpenAI's Agents SDK and Anthropic's Claude Agent SDK - orchestration models, tool ecosystems, sandboxing, and how to choose the right platform for your team.
9 min readTechnical content at the intersection of AI and development. Building with AI agents, Claude Code, and modern dev tools - then showing you exactly how it works.
What MCP servers are, how they work, and how to build your own in 5 minutes.
AI AgentsInstall Ollama and LM Studio, pull your first model, and run AI locally for coding, chat, and automation - with zero cloud dependency.
Getting StartedA practical walk-through of how to design, write, and ship a Claude Code skill - from choosing when to trigger, through allowed-tools, to the steps the agent will actually follow.
Getting Started
Getting Started with Code Llama 70B - Meta's Latest Release In this video, we explore the recently released Code Llama 70B from Meta that outperforms many other models across numerous benchmarks.I...

In this video I show you the new multimodal model by Meta AI that allows you to input speech or text and get a host of different outputs from speech-to-speech translation to speech-to-text...

The latest GPT Image 2 prompt-library repos are not just galleries. They point at a practical workflow for repeatable vi...

Inception Labs launched Mercury, the first commercial-grade diffusion large language model. It generates over 1,000 toke...

A practical comparison of OpenAI's Agents SDK and Anthropic's Claude Agent SDK - orchestration models, tool ecosystems,...

Codex can point at OpenAI-compatible model providers, local Ollama servers, and internal model proxies. Here is the prac...

OpenRouter Fusion turns multi-model panels into an API feature. The useful lesson is not to run every prompt through mor...

Anthropic's docs say the tokenizer introduced with Opus 4.7 can use up to 35% more tokens for the same text. Here is wha...

New tutorials, open-source projects, and deep dives on coding agents - delivered weekly.