Muse Image and Muse Video - New Generators from Meta
Meta has unveiled new image generators Muse Image and Muse Video, which integrate diffusion models and LLMs for creating graphics and videos.
Muse Image and Muse Video - New Generators from Meta
This is the first release after the rebranding of Meta GenAI to Meta Superintelligence lab. Muse Image and Muse Video are a sort of successors to the Emu models.
Essentially, these are something like GPT-Image/Nano Banana, where a diffusion model and LLM are tightly integrated (potentially sharing some layers). The model now writes Python code and parses the web for references, which was already present in Nano Banana.
They demonstrated how the model first writes code for generating a fractal and then clearly inserts it into an image. This is useful for infographics: now the model can first build a graph in Python based on your data and then place it in an image, which can be useful for restaurant menus where a clear layout is needed, which the model can create in conditional HTML.
Muse also corrects mistakes in the generated images (like Test-Time-Scaling, where multiple generations are made and the best one is chosen). Moreover, the last feature reportedly emerged during RL training.
In general, another Nano Banana, but with greater reasoning. On the t2i arena, it reached second place, narrowly surpassing NB2 and Reve-2, but the recently released Reve-2.1 has already overtaken Muse again.
On the Image Edit Arena - also second place, slightly ahead of MAI-image-2.5 (where part of my former team from Meta, where I used to work, went). The model is not yet present on other arenas.
There was some controversy. Someone thought it would be a good idea to generate faces of Instagram users simply by @-tag (this option is enabled by default), which not everyone appreciated. Honestly, I didn't expect that from Mark 😅.
The video model is currently without agents, but it confidently sits in the top 3 leaderboards, surpassing happyhorse from Alibaba.
Blog post - I think we can skip waiting for the paper. You can test the model on the arena, or for now only in the American Instagram - it takes ages for Meta to roll something out in Europe.
Why it matters
AnalysisThese new generators from Meta open up new possibilities for creating graphics and videos, which could significantly impact the content industry. The integration of AI into image generation processes makes them more accessible and efficient.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
Llama Family (Meta Llama)
A series of foundational open language models from Meta (Llama 3, 3.1, 3.3) that have become the industrial standard for the Open Weights ecosystem, local AI, and enterprise fine-tuning.
Four Resume Tips from OpenAI’s Head of Applied AI Development
Wulfie Bain, OpenAI’s head of applied AI development, shared four resume rules: keep it short, remove irrelevant details, highlight outcomes, and lead with the strongest achievements.
JSON — Structured Data for Programs, APIs, and AI Agents
A complete hands-on guide to JSON: syntax specifications, serialization via parse and stringify, REST API consumption, JSON Schema validation, and structured outputs for LLMs.
DOCX and XLSX Formats: When AI Needs to Deliver Production-Ready Documents
A comprehensive hands-on guide to generating and processing DOCX documents and XLSX workbooks with AI: comparing against Markdown and CSV, prompt engineering, formula design, and code automation.