Skip to content

PortraitGen

Edit 2D portrait videos through a 4D Gaussian field

Visit official site

What PortraitGen is

The tool is described as a 2D portrait video editor built on multimodal generative priors that lifts a portrait video into a 4D Gaussian field. It tracks SMPL-X coefficients and uses a neural Gaussian texture mechanism to generate and edit 3D portraits.

What you can do with it

  • Edit portraits with text-driven and image-driven instructions
  • Transfer the style of a reference image onto dataset frames
  • Change the clothing of the subject with AnyDoor
  • Adjust the lighting of video frames from a text description

Getting started

  • Visit the PortraitGen site
  • Choose text-driven editing, image-driven editing or relighting
  • Upload the 2D portrait video to edit
  • Enter a text instruction or pick a reference image

Example uses

  • A video editor generates a personalised portrait video
  • A game developer creates varied expressions and motions for characters
  • An animator works on character design and animation

Pros & cons

✓ What we like

  • An iterative dataset update strategy and a face-aware editing module are described
  • EMOCA latent expression space optimisation keeps expressions natural and consistent

! What to watch out for

  • Editing starts from a 2D portrait video lifted into a 4D Gaussian field
  • Face-aware editing relies on two editing passes for robustness

FAQ

What does the tool edit?

The material describes 2D portrait videos lifted into a 4D Gaussian field for multimodal portrait editing.

Which editing techniques are named?

InstructPix2Pix, neural style transfer, AnyDoor, IC-Light and EMOCA are named in the feature list.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review