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
Explore Muse Spark Open: an 816B open-weight multimodal model with a 1M-token window, built to run agentic video edits on your own Mac Studio setup.
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.
FLUX 3 Video Generator
Kling 3.0
Next-Gen AI Video Generator
Grok Video Generator
Create Videos from Text or Images with AI
MiniMax H3 video generator
MiniMax H3 AI Video Generator
Seedance 2.0
The Future of AI Video Is Here.
Inside Muse Spark Open: Open Weights at Desktop Scale
Shipped as downloadable checkpoints, this natively multimodal system handles coding, autonomous tooling and long-horizon jobs — and it is yours to self-host.
- Openly Released, Built in the USThis drop carries 816B total parameters with 42B active, a million-token window and native multimodality — the most capable open-weight launch yet from a US lab.
- Frontier Scale, Desk-Side FootprintTop-tier models no longer need a datacenter: pair two Apple M5 Ultra Mac Studios with the newest MLX runtime and the entire network runs under your own roof.
- Agentic Tasks, Already ProvenDriven through Muse Code, it landed genuine iOS patches and features in Ice Cubes, then operated DaVinci Resolve start to finish and re-framed the best shots vertically.
Three Ways to Deploy Muse Spark Open
Go local on Apple silicon, serve it with an engine you already trust, or point requests at a managed endpoint — every path gets you running quickly.
What Muse Spark Open Can Do
Open weights, a million-token window, true multimodality and workstation-grade memory bandwidth: the combination that lets long agentic video jobs finish on hardware you own.
One Million Token Context
Lengthy sessions stay coherent because the model can look back across a million tokens of earlier steps, plans and reasoning chains.
Coding That Runs for Hours
Built for engineering loops that span hours, it also drives desktop applications such as DaVinci Resolve from the first click to the final render.
Multimodal by Design
Images, footage and text enter the same model directly instead of passing through add-on adapters — the reason it can inspect a clip shot by shot.
Two Macs, One Cluster
Each M5 Ultra GPU can address up to 512 GB of unified memory, so a pair of Mac Studios pools roughly a terabyte and carries the full checkpoint as a single node.
1.2 TB/s Bandwidth
With 1.2 TB/s of throughput, the M5 Ultra feeds every per-token slice fast enough to keep interactive desktop editing responsive.
Safety Tuned Before Release
Extra post-training hardens the released weights; the model was measured against the Meta Superintelligence Scaling Framework and passed every threshold in the risk categories tested.
Muse Spark Open: Your Questions Answered
Straight answers on hardware requirements, licensing terms, download locations and hosted access for this open-weight model.
Can I run Muse Spark Open on my own machine, and what does it take?
Two Apple M5 Ultra Mac Studios running the newest MLX build. Each GPU addresses up to 512 GB of unified memory, so the pair offers roughly a terabyte for the full checkpoint.
How big is the model, and how much can it hold in context?
Expect 816B total parameters with 42B active at any moment, a 1M-token context window and native multimodality — sized for coding, agentic loops and long-horizon work.
Where do I get the weights, or call the model through an API?
Checkpoints live on Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models; managed endpoints are offered by Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI.
What runtimes and accelerators are supported?
Pick from MLX, llama.cpp, Unsloth, SGLang, vLLM, TensorRT-LLM or Transformers. All are tuned for NVIDIA, AMD, Intel and Apple silicon, with PyTorch underneath if you want to customize further.
Which license covers the released weights?
The Muse Model License governs use. Its limited commercial terms still leave most developers free to run these weights at no charge.
What safety checks happened before release?
The model was scored against the Meta Superintelligence Scaling Framework for the most severe risks, received extra safety post-training for open distribution, and cleared every release threshold — the model card has the details.
Start Building with Muse Spark Open
Pull the checkpoints from Hugging Face, route requests through a hosted inference partner, then share whatever you create in the Hugging Face Forums.