Inside WanGP
Create video, images, audio and 3D in a faster, connected workspace. Use the controls yourself, collaborate with Deepy, or connect your own AI agent. WanGP makes the most of your GPU while you check progress and launch jobs from another browser.
01 / MORE HARDWARE
WanGP was built for low VRAM. MMGP v4 adds major speed gains and further memory savings to that foundation, opening more room for demanding models, resolution and longer clips.
MMGP v4 adds speed and memory improvements to WanGP’s existing optimizations. v18 also accelerates LTX, Wan and HunyuanVideo VAE decoding and adds CUDA Graph acceleration for compatible workflows. Actual gains depend on the model, settings and hardware.
Faster generation · Efficient decoding · Low VRAMChoose compatible INT8, FP8, GGUF, NF4, NVFP4 or Nunchaku checkpoints to reduce memory use or take advantage of your GPU. Formats and acceleration options vary by model and hardware.
int8 · fp8 · GGUF · NF4 · NV FP4WanGP fetches compatible model files for your hardware. Optional preprocessing and postprocessing checkpoints download only when a workflow needs them, reducing unnecessary first-run downloads and disk use.
Models and tools arrive on demandThe project documents paths for older NVIDIA generations and current RTX hardware, plus dedicated AMD setup for supported RDNA 2, 3, 3.5 and 4 families.
NVIDIA + AMDBuild, edit, pause and save queues across video, image, audio and 3D. Check progress and add work from another connected browser, or run saved batches headlessly from the command line.
Save · resume · batchUse Deepy, the built-in agent, or connect your own through MCP. Agents can discover models, validate requests before generation, queue jobs and inspect images, video frames and 3D views. Run MCP alongside the browser to share the same queue, gallery and loaded models.
Bring your own agent · Or use DeepyChoose your workflow
Choose a model and quantization, start with a suitable memory profile, and scale resolution or duration as your hardware allows. WanGP fetches only the model and tool files the workflow needs, while its queue keeps larger projects moving.
Preprocessing, model-aware prompt enhancement, generation and postprocessing work together. Keep the assets you produce and reuse them across models, projects and future generations.
Every stage—from preparing an input to finishing the soundtrack— stays within reach of the same local workspace.
A faster UI, accessible from anywhere
Switch models and browse media in a more responsive interface with five themes, prompt dictation and floating Generate controls. From another connected device, watch live progress, inspect results and launch new generations. Galleries and queues stay synchronized; your unsent prompts stay local to each browser page.
Connect another device ↗Draw or expand masks, remove backgrounds, crop and resize references, extract pose, depth or optical flow, separate speakers, and remove noise or songs before generation.
Guide H3 and LTX 2.5 with references, start/end frames and sliding-window context. Use structured layout helpers for Ming Image or Ideogram, or edit reference images with Qwen Image 2.1. Dictate and review your prompt, then let the model-aware enhancer use the attached images and timing.
Carry generated images, clips, audio and rendered 3D views into new generations. Reuse references across compatible models, copy or move assets between projects, and keep your media and selections in persistent workspaces across restarts.
Upscale with SeedVR2 or LTX-2, restore detail with LTX 2.5, or smooth motion with RIFE and supported DLSS frame generation. Refine faces with H3, work with transparent layers from Ming Image or Qwen Layered, remove vocals, remux soundtracks and export lossless FLAC.
Deepy is WanGP’s built-in agent. Zero handles focused requests; Prime plans connected tasks and inspects intermediate results, including 3D views and animation frames. Save sessions, pause and resume, or continue in the phone-friendly Web App.
Explore Deepy & watch the demo →The plugin catalog spans galleries, finetune and LoRA management, image workbenches, prompt libraries, queue notifications, timeline tools and additional model or processor types.
Audio is first-class
Create songs, instrumentals, covers or score continuations with YuE2; turn a hum into a new song; or compose with MiniMax Music 3. Generate and edit speech with AuK, use up to three reference voices in compatible dialogue models, and finish tracks with vocal removal, mixing and lossless FLAC export.
Processing documentation
03 / 3D CREATION
Make assets and scenes, inspect them in the 3D viewer, then carry a new view into your next image or video.
Images to assets
Build textured PBR assets with TRELLIS.2 or Hunyuan3D 2.1, detailed meshes with TripoSG, Gaussian splats with TripoSplat, or separate parts with PartCrafter.
Photos to scenes
Work from photos or video with WorldMirror 2.0, MapAnything and Depth Anything 3 Points, or reconstruct from a single photo with MoGe-2.
Assets to new ideas
Rotate a 3D asset and save a view for image or video generation. Split meshes with GeoSAM2 or animate them with AnimateAnyMesh. Deepy can inspect angles and animation frames and prepare character sheets.
Ready to create