Multi-track video editor
Arrange generated or local clips on a timeline, add images and overlays, and lay in music or voice.
New
v3.8.6 adds Fibo 2.0
Your own AI studio.
Your own AI studio.
No cloud required.
Generate and edit images, video, music, speech and text with the newest open models — all running locally on your own GPU. Free, open source and private by design.
Runs the newest open models
- FLUX.3 Klein
- Qwen Image
- Z-Image
- Wan 2.2
- HiDream-O1
- Stable Diffusion 3.5
- Ideogram 4
- ACE-Step
- Whisper
- Gemma 4
- Krea2
- Chroma
- Qwen 3.8
- Kandinsky 5
- Fibo 2.0
- MOSS Image
- CogVideoX
- SkyReels V2
- Supertonic
- LongCat
Four studios, one app
From a single prompt to a finished piece.
Image, video, audio and text generation live side by side, with the editing tools to finish the job. Pick a workflow, choose a model and start creating.
Image studio
Generate from text, transform existing images, edit with plain-language instructions, inpaint, or sketch a rough idea and watch it become a finished piece.
- Text → Image
- Image → Image
- Image Edit
- Inpaint
- Paint → Image
- Upscale
- Extract
- Compose
72 image models · 35 LoRAs · 9 ControlNets
Video studio
Turn prompts or stills into motion, restyle existing footage, then upscale, interpolate and cut it all together in the built-in editor.
- Text → Video
- Image → Video
- Video → Video
- Frame → Frame
- Upscale
- Interpolate
- Extract
- Compose
24 video models. The clip here was made with MiniMax H3, sound included.
0:00 / 0:04
Audio studio
Compose full songs from a style prompt and lyrics, generate natural voices in 31 languages, and transcribe or translate speech.
- Text → Music
- Text → Speech
- Text → Audio
- Audio → Text
11 audio models. Press play to hear a track made with ACE-Step 1.5 XL Turbo.
0:001:00
Style prompt
“Melancholic synthwave combined with cyber-noir aesthetics. Driven by an analog Roland JU-06A synthesizer, driving 808 drum machine beats, and a pulsing, resonant bassline…”
60 s · 48 kHz stereo · 8 stepsMade in Amuse
Text studio
Chat with local language models that can see images, read documents and call tools like web search, a calculator or file search — all running on your own hardware.
- Converse
- Instruct
- Image → Text
- Documents
- Tool use
15 language models. This reply is from Qwen 3.8 27B with thinking enabled.
Can you help me plan a birthday party for my brother?
Thinking
The user is asking for help planning a birthday party for their brother. This is a general question, and I can help with it. I should ask some clarifying questions to make the plan more specific…
I'd love to help! To make this really useful, it would help to know a few things:
- How old is he turning? (A party for a 5-year-old looks very different from one for a 30-year-old or 60-year-old.)
- Rough budget?
- How many guests are you thinking?
- Any interests or themes? (sports, gaming, music, travel, food, etc.)
Built-in tools
Everything between the first draft and the final cut.
Professional editing and processing tools sit right next to the generators, so your work never has to leave the app.
Upscale images & video
Upscale up to 4× with 19 upscalers — Real-ESRGAN, SwinIR, Swin2SR, UltraSharp and more. Drag to compare a real before and after.
Frame interpolation
Multiply frame rates for silky-smooth playback or dramatic slow motion.
Inpaint & masking
Select, mask and repaint. Remove objects or fill areas with precise selection tools.
Feature extraction
Pull depth, edges, line art, pose or clean background cut-outs from images and video — ready to drive ControlNet.
LoRA & ControlNet
Steer style with 35 ready-made LoRAs, guide composition with 9 ControlNets, or bring your own.
Voice & speech
Lifelike voices in 31 languages with Supertonic, plus transcription and translation with Whisper.
An assistant with tools
Local LLMs that read your documents, look things up, do the math and see what's on screen.
Media gallery
Every output organized in one place. Filter by type, revisit settings and pick up where you left off.
Model manager
One-click downloads with a queued download manager and in-app updates.
Quality modes
Draft (int4), Standard (fp8) or Production (bf16) — trade VRAM for fidelity.
Smart memory
Offload weights across GPU and CPU to fit big models on smaller cards.
Batch automation
Run prompts from files or folders and process media in bulk.
Live step preview
Watch every diffusion step take shape while it generates.
Bring your own models
Add checkpoints, LoRAs and ControlNets in Safetensors, GGUF or ONNX.
Automatic updates
Get notified in-app the moment a new version ships.
English & 日本語
A fully localized interface in English and Japanese.
Private by design
Your ideas never leave your desk.
Amuse runs every model on your own hardware. Prompts, images, recordings and documents stay on your PC — nothing is uploaded, metered or stored on someone else's server.
Works offline
Once a model is downloaded, generating needs no connection.
No account
No sign-up and no login to use Amuse.
No credits
No subscriptions or usage caps. Create as much as your GPU can handle.
Your folders
Outputs are saved to the folders you choose.
Model library
The newest open models, one click away.
A curated catalog of nearly 200 models and add-ons, preconfigured to run well on consumer GPUs. New pipelines arrive in almost every release.
- Image models
- 72
- Video models
- 24
- Audio models
- 11
- Language models
- 15
35 LoRA adapters19 upscalers12 extractors9 ControlNets
No models match “”.
Models download directly from their publishers on Hugging Face. Each model is subject to its own license.
Hardware
Built for the GPU you already have.
Choose the backend that fits your card. Amuse installs and manages each runtime environment for you — no Python wrangling required.
NVIDIA
- CUDA 13.0
- Vulkan
Native CUDA acceleration. RTX cards are recommended — their Tensor Cores are built for AI workloads.
Requires driver 580.65 or later
AMD
- Vulkan
Radeon GPUs run on the cross-vendor Vulkan backend, powered by the lightweight stable-diffusion.cpp engine.
Support varies by GPU and driver
Intel
- Vulkan
Intel Arc and other Vulkan-capable GPUs use the same cross-vendor Vulkan backend.
Depends on your Vulkan driver
Three engines under the hood
-
Diffusers & Transformers
Hugging Face pipelines on PyTorch for image, video, music and language models in Safetensors.
-
stable-diffusion.cpp
A lean C++ engine for GGUF-quantized models — fast, light and friendly to smaller GPUs.
-
ONNX Runtime
Optimized ONNX models for speech, voice, upscaling and feature extraction.
Pick your balance
ModePrecisionVRAMDetail
Draftint4LowestGood
Standardfp8ModerateNear-identical
Productionbf16HighestMaximum
For most people, Standard mode is indistinguishable from Production — with significantly lower memory use.
Windows 10 (1809+) or Windows 11, 64-bit. The Vulkan backend runs models tagged sd.cpp in the model library. Actual compatibility and performance depend on your GPU model, driver version and backend.
Open source
Built in the open. Shipped constantly.
Amuse is Apache 2.0 licensed and built on .NET 10. Read the code, report issues, suggest features, or build it yourself.
Build from source
# 1. Clone the repository
git clone https://github.com/saddam213/AmuseAI.git
# 2. Open Amuse.sln in Visual Studio 2026
# (.NET desktop development workload)
# 3. Restore packages, then build
# Build → Build Solution (Ctrl+Shift+B)
# 4. Run Amuse.exe from the build outputBuilding from source requires a Font Awesome Pro v7 license for the interface icon fonts.
Recent releases
All releases-
Fibo 2.0
Adds support for the Fibo 2.0 image pipeline — Dev and Edit Turbo.
-
MOSS Image
Adds support for the MOSS Image pipeline — SFT and Base.
-
FLUX.3 Klein 6B
Adds support for the FLUX.3 Klein pipeline — 6B Base and 6B Distilled.
- Stars
- 663
- Forks
- 113
- License
- Apache 2.0
Made possible by these open-source projects
FAQ
Questions, answered.
Can't find what you're looking for? Ask on GitHub.
Is Amuse really free?
Yes. Amuse is free and open source under the Apache 2.0 license. Compiled releases are distributed under the Amuse 2.0 License, because they bundle third-party software and dependencies that come with their own licensing terms.
Does it need an internet connection?
Only to download models, runtime environments and updates — and for optional online tools such as web search. Generation itself runs entirely on your PC.
What hardware do I need?
A 64-bit Windows 10 (1809 or later) or Windows 11 PC with a dedicated GPU. NVIDIA RTX cards give the best performance through CUDA; AMD and Intel GPUs run through Vulkan. Memory requirements vary by model — quality modes, memory offloading and GGUF models help bigger models fit on smaller cards.
Which models can I use?
Amuse ships with a curated catalog that includes FLUX, Qwen Image, Z-Image, Stable Diffusion, Wan, ACE-Step, Whisper, Qwen and Gemma, among many others. You can also add your own Safetensors, GGUF and ONNX models, LoRAs and ControlNets.
Can I use what I create commercially?
That depends on the model. Each model in the catalog is published under its own license, so check the terms on its Hugging Face page before using outputs commercially.
Is there a macOS or Linux version?
Not currently. Amuse is a native Windows application built with .NET and WPF.
Environment setup failed — what should I try?
On Windows 11, creating an environment can require elevated permissions. Close Amuse, right-click it, choose Run as administrator and try again.
How do I report a bug or request a feature?
Open an issue on GitHub using the bug report or feature request template. Contributions of every kind are welcome.
Start creating. Locally.
Download Amuse for free and turn your GPU into a complete creative studio.