Video
Generate from text, animate an image, continue a clip or guide motion with reference media.
Video · Image · Audio · 3D · Worlds
One Generative Platform.Infinite Possibilities.
Generate video, images, audio, 3D—and worlds you can explore.
Your full production pipeline, from first input to reusable assets.
Hundreds of models. For humans and AI agents.
Create hands-on, bring your own agent, or use Deepy, our built-in agent.
- Super fast
- Low VRAM / GPU Poor
- Free to run locally
Media Generator · Your local creative workspace
Create directly. Control every detail.
Video, images, audio and 3D.
Start with a prompt or bring your own references. Choose a model, shape the result with generation controls, and keep your creations together in the gallery. Everything runs on your own computer.
Explore all generation tools- Choose a model and references
- Generate and inspect the result
- Reuse it in the next medium
Images
Create illustrations and photos, edit with references, and refine selected areas with masks.
Audio
Make songs, instrumental music, voices and sound for your next creation.
3D
Turn images into 3D assets, inspect them in the viewer and render a new angle.
Meet Deepy. Your built-in agent.
Describe what you want to create. Deepy plans the steps, uses WanGP’s models and tools, and checks intermediate results with you. Continue in the full workspace or the phone-friendly Deepy Web App.
Explore Deepy and watch the demo- Describe the result and share your references
- Let Deepy coordinate the models and tools
- Review, steer and refine together
Bring the agent you already use.
Connect your preferred MCP-compatible agent to WanGP. It can discover models, validate settings, queue jobs and inspect outputs alongside your browser workspace.
Connect through MCP- Connect your agent through MCP
- Discover models and validate requests
- Generate, inspect and deliver the results
Built for agents
Tools that explain themselves.
Results your agent can see.
Model discovery, clear input requirements and preflight checks help agents prepare valid requests. Images, sampled video frames and rendered 3D views let them inspect the results and choose the next step.
Use the browser and MCP together
python wgp.py --mcp-webuiConnect your agent to
http://127.0.0.1:8000/mcpOne running WanGP instance shares the queue, gallery and loaded models with your browser.
Models you can run in WanGP
Explore the model guideWorlds
Generate a place. Walk through it.
- Hy-Nomas 8B
- EVOKE Turbo 14B
A selection of supported models and variants, from media generation to interactive world exploration.
Audio is first-class
Give your ideas
a voice. And a soundtrack.
Create songs with YuE2 and ACE-Step, compose with MiniMax Music 3, or generate speech with Qwen3 TTS. Work on music and voice in the same platform you use for images, video and 3D.
Explore the creative toolbox
Asset reuse in practice
One image.
A whole new dimension.
Build on earlier generations: turn an image from your gallery into a 3D asset, render a new angle and use that view to begin a video. Each step carries your creative work forward.
Explore 3D workflows- 01 / IMAGE
Make a character
Generate an image or bring your own reference.
- 02 / 3D
Give it form
Build a mesh or textured asset with a compatible 3D model.
- 03 / VIEW
Find your angle
Rotate the asset in the 3D viewer and save a new view.
- 04 / VIDEO
Set it in motion
Use the rendered image as the starting point for your next shot.
Interactive world models
Create a world.
Step inside.
Start with a description or an image, then explore with the keyboard. The Worlds gallery keeps your journey ready to continue, record as video, or capture as a new view for another generation.
Start with 8 GB of VRAM. Exploration speed depends on the model and GPU; near-real-time movement needs fast, recent hardware.
Hy-Nomas 8B
Describe a place and move through it. Earlier views help keep locations and characters consistent. Add an event, turn the camera, and choose how much history the model remembers.
EVOKE Turbo 14B
Bring an image of a street, landscape or interior. Scene-depth memory helps preserve the layout when you return. Explore at 640 × 384, with more time needed for each move.
01 Generate a starting scene02 Explore from the keyboard03 Save, record, or capture a view
See real demos & learn to explore →
The full production pipeline
From first input to finished work.
And your next generation.
Preprocess your inputs, enhance your prompts, generate and finish your media. Keep the assets you create ready to use across models, projects and future generations.
- 01
Preprocess
Prepare references with masks, background removal, cropping, pose and depth extraction, or audio cleanup.
- 02
Enhance prompts
Shape your idea with model-aware prompt enhancement that uses your references, timing and generation settings.
- 03
Generate
Create video, images, audio, 3D and worlds. Queue jobs, guide the model and inspect results in one workspace.
- 04
Postprocess
Upscale, refine details, interpolate frames, mix audio or replace a soundtrack with the built-in finishing tools.
- 05
Reuse assets
Bring generated images, clips, audio and rendered 3D views into new jobs. Build on earlier generations across models and media.
Hands-on creation
You create.
WanGP does the heavy lifting.
Prepare inputs, enhance prompts, choose your models and finish the results. Your tools and reusable assets stay together in one browser workspace.
Explore the platformCreation with an agent
Bring your own agent.
Or use Deepy.
Connect your preferred agent through MCP, or work with Deepy, WanGP’s built-in agent. Let an agent coordinate preparation, generation and finishing while you steer the result.
Meet DeepyOne runtime · Shared queue · Shared galleryBrowser + MCP
Deepy Games · In development
Imagine it.
Then play it.
Made in one shot by Deepy, 100% locally in WanGP: code, artwork, sound and playable games from one request. A few small manual fixes refine these showcase examples.
Explore 10 generated gamesPlay the browser demos. Download the games for WanGP.
One shot. 100% local. Made by Deepy.
More from the GPU you already own
Super fast. Low VRAM.
Built to keep you creating.
Already optimized. Still getting faster.
MMGP v4 helps fit larger work into less memory. Recent releases improve older RTX support, reference-based generation and local language-model speed. Experimental Profile 0 automatically shares RAM and spare VRAM between the models you use.
Room for more ambitious ideas.
Memory profiles, quantized checkpoints and on-demand model downloads help fit the work to your machine. Select models start at 6 GB of VRAM; requirements depend on the model, resolution and settings.
Your workspace, wherever you are.
Watch progress and launch the next generation from another connected browser. Shared queues and galleries keep you in step while your computer does the generation.
GET STARTED / INSTALL
Choose your way in.
Four installation paths. Choose the level of guidance and control that fits you.
01Official GitHub
Project-maintained scripts
Official scripts
Clone the official repository, then use its installer to create and manage an isolated environment suited to your hardware.
- Windows, Linux and macOS scripts
- Auto or guided manual setup
- Matching run, update and manage scripts
scripts\install.bat./scripts/install.sh02App manager
One-click app manager
Pinokio
Install and run WanGP through Pinokio. The WanGP README recommends Morpheus's community scripts wan2gp and wan2gp-amd.
- Graphical install and launch flow
- Separate NVIDIA and AMD scripts
- Update from Pinokio
03Windows launcher
All-in-one desktop launcher
Wan2GP Desktop
Install, update and launch WanGP from one window. Hardware detection and Auto-Tune help select the right environment and memory profile.
- Windows 10 / 11
- Prerequisites handled for you
- Built-in launch and maintenance tools
04Official · Windows x64
Guided desktop installation
WanGP Manager
Choose a folder, a WanGP version and the recommended profile for your GPU. Manager installs Python and dependencies, then launches your workspace.
- Guided setup and hardware detection
- Launch in an app window or browser
- Update, switch versions and roll back
Experimental. The download becomes available when WanGPManager.exe is published in the latest release.
Need GPU-specific details? Use the official guides before changing Python, PyTorch, CUDA or ROCm versions.
QUESTIONS / FAQ
Good to know before you generate.
Practical answers drawn from the project documentation and the questions that recur in the WanGP community.
01What is WanGP?
WanGP is one generative platform for video, images, audio, 3D and explorable worlds. Run models on your own GPU, create in the browser, work with Deepy, or connect an agent through MCP. Deepy Games is a new capability in development. WanGP is free to run locally under its Community License.
02Are Wan2GP and WanGP the same project?
Yes. Wan2GP is the original project name; WanGP is the current name. Older guides, folders, scripts and community posts may still say Wan2GP. The former repository address redirects to the official WanGP repository.
03How much does WanGP cost?
WanGP is free to install and run locally, including for internal company use. The official project will never ask for a license fee, subscription or donation simply to run WanGP on your own computer. Paid hosting, SaaS, white-labeling and other restricted commercialization require separate permission; model and third-party component licenses also continue to apply. Read the WanGP Community License 2.0
04How much VRAM do I need?
Select smaller models can run with as little as 6 GB of VRAM. Larger models, higher resolutions and longer videos require more. The practical requirement depends on the chosen model, quantization, memory profile, frame count and resolution—so start with a recommended low-VRAM model and scale up.
05Which GPUs are supported?
WanGP supports NVIDIA GTX 10-series, RTX 20-series and newer GPUs, with generation options tailored to each generation. AMD support covers several RDNA 2, RDNA 3, RDNA 3.5 and RDNA 4 families; follow the dedicated AMD guide because setup and compatibility vary by GPU architecture and operating system.
06Which installation option should I choose?
Choose one of four paths. First: the scripts in the official GitHub repository for the most direct, project-maintained setup on Windows, Linux or macOS. Second: Pinokio for a graphical app-manager workflow. Third: Wan2GP Desktop for an all-in-one Windows launcher with hardware detection and maintenance tools. Fourth: the experimental official WanGP Manager for guided Windows installation, launching, updates and rollback.
07Does WanGP download every model during installation?
No. WanGP downloads model checkpoints and optional preprocessing or postprocessing assets only when a requested workflow needs them. First use can involve a substantial download, so leave enough disk space and allow it to finish.
08Can WanGP run offline?
Generation runs locally. Download the models and dependencies you need, then launch with --offline to prevent downloads and report any missing files. Optional Deepy web search and remote assistant engines need internet access.
09What should I do if I run out of memory?
Reduce resolution or frame count, choose a smaller or quantized model, lower the batch size, and use a more memory-efficient WanGP profile. Closing GPU-accelerated apps also helps. The repository includes scripts that launch Chrome without GPU acceleration, which can return additional VRAM to WanGP.
10Why is generation slower than expected?
Speed depends on the model, GPU, attention backend, quantization, memory profile, resolution and number of steps. Use the hardware-appropriate installer settings first. Optional accelerators such as Triton and SageAttention can help on compatible GPUs; the official installation guide documents the supported combinations.
11How do I update WanGP?
Keep using the same path you installed with. Official-script installations use scripts/update.bat on Windows or scripts/update.sh on Linux/macOS. Pinokio installations update through Pinokio. Wan2GP Desktop installations use the launcher’s built-in maintenance tools.
12Where can I get help?
Join the official Discord community for setup help, workflow ideas and sharing generations. For reproducible bugs, search existing GitHub issues first and open a new issue with your operating system, GPU, VRAM, installation method, model, settings and terminal error.
13Can I check progress and launch jobs from my phone?
Yes. Connect to the same running WanGP server from another browser or use the phone-friendly Deepy Web App. Galleries, progress and the queue stay synchronized, and you can launch new generations. WanGP must remain running on your computer; use the network access guide to configure remote access.
14Can I use my own AI agent instead of Deepy?
Yes. Connect an MCP-compatible agent to discover models, check inputs, queue jobs and inspect results. Launch WanGP with --mcp-webui to share the running instance, queue and gallery with the browser. Deepy is the built-in option if you prefer to start with an agent already integrated into WanGP. Read the MCP setup guide.
Built in the open
Generate locally. Learn together.
WanGP grows through open development and a community of creators, tinkerers and contributors.
Official project
This is the official WanGP website. WanGP is free to install and run locally—no license fee, subscription or donation is required. See the license terms.
Stay safe
Use the official GitHub repository linked here. WanGP is not affiliated with third-party services using similar names unless the repository explicitly says otherwise.