
None
None
Long Story Video Skill
Turn your idea into 10s–10min video—AI writes scripts, prompts, and generates footage automatically.

Ads Video Skill
Generate professional ads and sales videos—AI auto-generates scripts, prompts, and footage.
3D Science Explainer Video Skill
Convert scientific concepts into stunning 3D explain animations
Feedback
freeTrialImage.bannerPity
freeTrialImage.upgradeUnlock
- ✓freeTrialImage.benefitHd
- ✓freeTrialImage.benefitWatermark
- ✓freeTrialImage.benefitUnlimited
Muse Spark Open
Run Muse Spark Open on your own Mac: 816B open-weight parameters, 42B active, a 1M-token window and native multimodal reasoning for agentic edits.
All Tools
Discover our comprehensive AI-powered animation toolkit
MiniMax H3
MiniMax H3 AI Video Generator
Seedance 2.5
The Future of AI Video Is Here.

Seedance 2.0
The Future of AI Video Is Here.

Seedance 2.0
The Future of AI Video Is Here.

Gemini Omni
Gemini Omni Video Generator
Image to Video
Generate videos from images

Gemini Omni 1.2 Flash

AI Multi-Scene Shorts Generator
Create viral AI Shorts instantly
Take Control with Muse Spark Open
Shipped as downloadable weights, this natively multimodal engine handles coding, autonomous agent loops and long-horizon jobs entirely on hardware you own.
- Open Weights, Built in the USThe strongest open-weight release yet from a US lab arrives as a downloadable checkpoint — 816B total parameters, 42B active, a million-token window and multimodality baked in.
- Datacenter-Class Power on a DeskNo locked-away cluster required: pair two Apple M5 Ultra Mac Studios with the latest MLX runtime and the entire model runs on hardware sitting right beside you.
- Agentic Work Verified On-DeviceMuse Code already pushed genuine iOS fixes and features into Ice Cubes, then drove DaVinci Resolve start to finish and reframed the best shots into vertical cuts.
Three Ways to Deploy Muse Spark Open
Pick the route that suits you — local Apple silicon, an inference engine of your choosing, or a managed hosted endpoint.
What Muse Spark Open Can Do
A million-token window, native multimodality, desktop-grade memory bandwidth and downloadable weights — the ingredients behind dependable agentic video editing.
One Million Token Context
With a 1M-token window, Muse Spark Open keeps long agent sessions and multi-step reasoning chains anchored to decisions made much earlier.
Long-Horizon Agentic Coding
Built for software engineering loops, the model sustains lengthy jobs and drives tools such as DaVinci Resolve from first step to final export.
Multimodality Without Adapters
Mixed inputs are read natively instead of through bolted-on layers — which is precisely what makes frame-by-frame footage review workable.
Two Mac Studios, One Supercomputer
Unified memory lets each M5 Ultra GPU address up to 512 GB, so a pair of Mac Studios holds the full model as a single machine with a terabyte pooled.
1.2 TB/s Memory Bandwidth
The M5 Ultra pushes 1.2 TB/s, enough to feed each small per-token slice and keep inference snappy on a desktop.
Safety Post-Training Included
Open weights got extra safety post-training: the model was measured against the Meta Superintelligence Scaling Framework and passed its release bar across the tested risk categories.
Muse Spark Open Questions, Answered
Answers on hardware requirements, licensing, downloads and running Muse Spark Open.
Which hardware do I need for local Muse Spark Open inference?
Two Apple M5 Ultra Mac Studios running the newest MLX build — each addresses up to 512 GB of unified memory, giving roughly a terabyte for the full model.
How big is the model, and what context length does it support?
816B total parameters with 42B active, a 1M-token context window and native multimodality, tuned for coding, agentic loops and long-horizon work.
Where do I download the weights or call the model as an API?
Weights are available on Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry Models; hosted APIs run at Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI.
Which inference engines and chip families work?
Pick from MLX, llama.cpp, Unsloth, SGLang, vLLM, TensorRT-LLM or Transformers — tuned for NVIDIA, AMD, Intel and Apple silicon, with PyTorch underneath for deeper customisation.
What license covers the released weights?
The Muse Model License applies; its limited commercial terms still leave broad room for developers to run the model at no cost.
How was safety evaluated before launch?
It was scored against the Meta Superintelligence Scaling Framework across the most severe risk areas, received extra safety post-training for open release, and cleared the thresholds — the model card has the details.
Start Building with Muse Spark Open
Pull Muse Spark Open from Hugging Face, call it through hosted inference partners, and share what you create on the Hugging Face Forums.
