New animation
Add a character and a motion reference.
3D character
Animate to track your reference. You can then play or scrub the movement here.
Preview
Your video will appear here
Add your files, then animate.
Loading pose model
Preparing the browser motion tracker.
Library
Videos saved in this browser.
Download a copy before clearing browser data.
No saved videos yet.
Your exported videos will appear here.
About
Motion transfer in your browser.
How it works
In Image mode, Flowmotion detects the character's joints and moves sections of the image to follow your reference video. In 3D character mode, a full-body image can be used to build a textured, rigged character with estimated depth. You can also load an existing humanoid GLB or VRM model. Inspect the model from different angles and choose the camera view for export.
Generate sides and back uses local AI texture completion and downloads about 230 MB of model and runtime files when first used. Turn it off to use simple color estimates. Failed AI generation keeps your previous character; it does not silently switch to simple textures.
All processing happens in your browser. Your images, models and videos stay on your device. No GPU server or API key is required.
Exports are silent H.264 MP4 videos with no added watermark. You can also download the 3D character as a rigged GLB. The library uses this browser's storage; download a copy of any video you want to keep.
Limitations
Image mode is 2D cutout animation. It cannot create unseen sides of the character, reconstruct hidden limbs, or change facial expressions. Large turns and overlapping limbs may look distorted.
A 3D character built from one image has estimated depth. Unseen sides, the back and hidden body parts are approximated; the result is not an accurate 3D scan. Use a front-facing, full-body image with arms and legs separated. Portraits and strongly overlapping limbs are unsuitable.
Existing 3D models need a humanoid skeleton, skin weights and embedded textures. Body tracking estimates motion from one camera; occlusions, fast turns and depth can be inaccurate. Facial expressions and detailed finger motion are not transferred.
For 2D images, transparent PNGs give the cleanest edges. Select Adjust joints if automatic detection misses the pose. For either mode, use a reference showing one person clearly.
MP4, MOV, MKV, AVI, WebM, HEVC/H.265 and other common video formats can be imported. Files your browser cannot play are converted locally to an editing copy of up to 10 seconds. Conversion can take longer for large or high-resolution videos.
The Reference video background is available in Image mode and keeps the original person visible behind your character. 3D mode uses a flat background.