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
Muse Spark Open puts a self-hosted, million-token multimodal model behind your video edits — free weights, no cloud lock-in.
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.
Veo3.1
Create Stunning Videos with Veo3.1
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
What the Open Release Puts in Your Hands
The checkpoints are public: a natively multimodal engine built for coding, autonomous workflows and long-horizon jobs that you can host entirely on your own hardware.
- Public Weights, Built in the USThis drop is the strongest openly available checkpoint to come out of the US — 816B parameters in total with 42B active per token, a million-token window and native multimodality.
- Frontier Power at Your DeskFrontier-class capability should not sit behind a datacenter door; pair a couple of Apple M5 Ultra Mac Studios with the latest MLX build and the model boots right where you work.
- Agentic Runs, Already ProvenMuse Code drove real iOS fixes and features into Ice Cubes, then steered DaVinci Resolve end to end — auto-cropping the strongest frames into vertical cuts.
Three Ways to Get Muse Spark Open Running
Choose your path: serve the weights on local Apple silicon, wire up the inference engine you prefer, or send requests to a hosted endpoint.
Inside Muse Spark Open
A million-token window, native multimodality, desktop-grade memory bandwidth and openly published weights — the ingredients that make agentic video work feasible.
A Million Tokens of Memory
With a 1M-token window, drawn-out agentic sessions and multi-step reasoning chains keep every earlier decision in view.
Agentic Coding That Goes the Distance
Built for software engineering and autonomous loops, Muse Spark Open sustains long-horizon jobs and steers end-to-end tools such as DaVinci Resolve.
Multimodal by Design
Mixed inputs are read natively rather than through bolted-on adapters — which is exactly what enables frame-by-frame footage review.
A Pair of Mac Studios, One Rig
Unified memory lets each M5 Ultra GPU address up to 512 GB, so two Mac Studios hold the entire model as a single machine with a terabyte of pooled memory.
1.2 TB/s of Bandwidth
The M5 Ultra shifts 1.2 TB/s — plenty to feed the small per-token slice and keep responses snappy on a desktop.
Safety Tuned Before Release
Open weights receive extra safety post-training: the model was scored against the Meta Superintelligence Scaling Framework and cleared its bar across every measured risk category.
Muse Spark Open Questions, Answered
Straight answers on hardware, licensing, downloads and deployment for Muse Spark Open.
What kind of machine does Muse Spark Open need to run locally?
Two Apple M5 Ultra Mac Studios on the latest MLX build — each addresses up to 512 GB of unified memory, giving roughly a terabyte together for the full model.
How big is the model, and how much context can it hold?
816B parameters in total with 42B active per token, a 1M-token context window and native multimodality — aimed at coding, autonomous work and long-horizon jobs.
Where do I get the weights, or call the model as an API?
Checkpoints are available on Amazon SageMaker AI, Google Cloud Model Garden, Hugging Face and Microsoft Foundry Models; hosted endpoints run at Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI.
Which inference engines and chips does it support?
Pick from MLX, llama.cpp, Unsloth, SGLang, vLLM, TensorRT-LLM or Transformers — tuned for NVIDIA, AMD, Intel and Apple silicon, with PyTorch underneath when you want to dig deeper.
Which license covers the released weights?
The Muse Model License applies; its limited commercial terms still leave broad room for developers to run Muse Spark Open at no cost.
How was safety checked before the release?
It was scored against the Meta Superintelligence Scaling Framework for the most severe risks, received extra safety post-training for open weights, and passed the release thresholds — the model card holds the details.
Start Building with Muse Spark Open
Pull the checkpoints from Hugging Face, call the model through hosted inference partners, and share what you create on the Hugging Face Forums.