Meta introduced Meta Muse on July 7, 2026, and Meta Muse arrived as two separate models at once: Muse Image, a fully available AI image generator, and Muse Video, an early preview of an AI video generator. Both come out of Meta Superintelligence Labs, the research group Meta built to chase frontier AI work, and Meta Muse is that group's first real swing at image and video generation. Muse Image rolled out broadly within days, while Muse Video is still limited to a small group of creators. You can read Meta's own announcement in Introducing Muse Image and Muse Video, and this guide walks through what Muse Image and Muse Video actually do, how the Meta Muse launch played out, and where each model stands as of today.
Meta Muse is Meta's name for its first in-house family of media generation models, and it exists because Meta wanted an image and video engine it fully controls instead of leaning on outside vendors. Before this release, Meta AI leaned on licensed and partner models for image generation. Muse Image and Muse Video change that by giving Meta Superintelligence Labs its own foundation to build on, and Meta has already framed the Muse family as the base for future creative tools across its apps. Muse Image is generally available, while Muse Video is a preview meant for creators who want an early look at where this technology is headed next. Meta describes Muse Image and Muse Video as reasoning-driven rather than purely generative, meaning neither model just draws or animates a prompt in one blind pass. Instead, the model studies the request, plans out the result, and refines it before showing you anything, which is a meaningful shift from how earlier Meta AI image tools worked. Understanding this distinction matters because it explains why Meta Muse behaves more like a careful collaborator than a one-shot generator, and it sets up everything else this guide covers about Muse Image and Muse Video.
Muse Image is the flagship half of Meta Muse, and it covers text-to-image generation, single-image editing, and multi-image editing in one model. Anyone comparing image generators today should know that Meta Muse packs all three of those jobs into a single model instead of splitting them across separate tools. Type a plain-language prompt and Muse Image generates a picture from scratch, or hand it an existing photo and ask Muse Image to change one object, swap a background, or blend elements from several images together. What makes Muse Image stand out is agentic tool use. Rather than only relying on what it learned during training, Muse Image can reach out to web search and coding tools mid-generation to check a fact, pull a reference, or work out a layout detail before finishing an image. The model also relies on emergent self-refinement, where Muse Image spends extra compute internally reviewing and improving a draft before it ever reaches your screen, similar to how a human illustrator might sketch, step back, and redo a section that is not working. On top of the core model, Muse Image ships with ready-made presets: restore an old, faded photograph, try out a new hairstyle on a selfie, redesign a room's furniture and colors, push an image into a different art style, or generate a graphic that contains crisp, readable text. Each preset is a direct product of the same reasoning approach behind Muse Image, since the model plans the edit before applying it rather than guessing pixel by pixel. This preset library is one of the fastest ways to feel the difference Meta Muse makes over older, simpler image tools.
Meta backed up Muse Image with real benchmark numbers instead of vague marketing claims, and those numbers are a big part of why Meta Muse is being taken seriously by people who track image generation models closely. On Arena, the crowdsourced leaderboard that ranks models by human preference, Muse Image holds the number two spot for text-to-image generation, for single-image editing, and for multi-image editing. Muse Image's Arena Elo score for text-to-image generation sits at 1,280, just behind OpenAI's GPT Image 2 at 1,385, and comfortably ahead of most other image generators on the board. On editing tasks specifically, Muse Image beats Google's Nano Banana 2 while still trailing GPT Image 2 overall. Those numbers matter for anyone deciding whether to build on Muse Image instead of a competing tool, because Arena rankings reflect thousands of blind, head-to-head human votes rather than a single lab's internal testing. Muse Image landing at number two on a leaderboard this competitive, on its very first public try, is a strong signal that Meta Superintelligence Labs closed most of the gap with more established image generators in a short amount of time. It also means Muse Image is not a toy feature bolted onto Meta AI for novelty; the model is being positioned as a serious contender against the biggest names in image generation, and Meta Muse now has the scores to back that positioning up.
Muse Video is the second half of Meta Muse, and it is still an early preview rather than a full public release. Pairing Muse Video with Muse Image under one Meta Muse brand signals that Meta wants people to think of image and video as a single connected toolkit rather than two separate products. Meta has said Muse Video is "coming soon to creators and in Meta AI," without committing to an exact date, so the video side of Meta Muse is clearly earlier in its rollout than Muse Image. Muse Video shares its pretraining foundation with Muse Image, which is part of why Meta treats image and video as one connected product line rather than two unrelated tools. The standout feature of Muse Video is native audio support. Instead of generating a silent clip and bolting on music or a voiceover afterward, Muse Video produces synchronized audio in the same generation pass as the video itself, so dialogue, sound effects, and ambient noise line up with the motion on screen from the start. On Arena's text-to-video leaderboard, Muse Video already ranks third by human preference, which is a solid debut for a model still labeled a preview. Meta has been upfront that Muse Video has real gaps left to close, specifically around audio-video synchronization in tricky scenes and rendering physically accurate fast motion, so this half of Meta Muse is expected to keep improving before a full release. Meta is aiming Muse Video at creators, filmmakers, and small teams who want short-form video ads, product teasers, and social clips with cinematic visual quality and audio that actually matches the footage on screen.
The Muse Image launch did not stay smooth for long. Within days of going live, Meta faced heavy criticism because Instagram had automatically opted public accounts into a feature that let other people remix someone's public photos into new AI images using Muse Image, without notifying the person whose photo was used. Critics, including the actors' union SAG-AFTRA, pushed Meta to make the feature opt-in instead of opt-out, warning that reusing public images this way opened the door to harassment and impersonation. The episode showed that rolling out Meta Muse across apps as large as Instagram carries real consent questions that a benchmark score alone cannot answer. Meta responded quickly: on July 10, 2026, the company pulled the cross-account remix feature from Instagram entirely and admitted it "missed the mark" on privacy. Importantly, this rollback only affected that one Instagram capability. Muse Image itself remained available through the Meta AI app and WhatsApp the whole time, and Instagram's other Meta Muse features kept working normally. If you are trying Muse Image today, this episode is worth knowing about, since it shaped how Meta is now rolling out Muse Image and will likely shape how Muse Video launches on Instagram down the road.
Availability differs by app right now, and it is worth knowing exactly where each model of Meta Muse currently works, since the Meta Muse rollout has already changed once and could keep shifting as Meta responds to feedback. Muse Image is live inside the Meta AI app and on the meta.ai website, and inside WhatsApp chats in a limited set of countries, with broader availability expected in the coming months. On Instagram, Muse Image still powers Stories creation tools in the United States, though the cross-account remix capability described above is currently switched off while Meta reworks its consent controls. Muse Video, being an earlier-stage preview, is not broadly self-serve yet outside the creator group Meta invited early. On pricing, Meta keeps the entry point simple: basic access to Muse Image through Meta AI is free, so anyone who already uses Meta AI, Instagram, or WhatsApp can try it without paying anything upfront. If you generate a lot of images or need higher limits, heavier usage of Muse Image is gated behind Meta's monthly subscription plans, which unlock expanded generation capacity and extra features on top of the free tier. To actually use it, open the Meta AI app or meta.ai in a browser, type a plain description of the image you want, and Muse Image will generate a picture along with a few variations to choose from. To edit an existing photo instead, upload it and describe the change in plain language, such as swapping the background or trying a new hairstyle, and Muse Image applies the edit while keeping the rest of the photo intact. As Meta keeps expanding this rollout, expect Muse Image's presets and Muse Video's creator tools to show up in more places across Meta's apps, with Meta Muse likely picking up new consent controls and regional availability at the same time.