LivePortrait

LivePortrait turns still portraits into motion-led videos and suits technically confident creators who want local control over facial animation. Its flexibility and free MIT-licensed implementation are strong draws, but setup takes several dependencies and commercial users must replace the included InsightFace detection models.

Overview

LivePortrait can animate a source image from a driving video, audio, text, or generation, and can edit portrait video by combining a source video with a driving video. It focuses on face animation rather than depth effects, with MP4 as its stated export format.

The project provides local inference code, pretrained weights, and a Gradio interface, as well as an online demo hosted on Hugging Face Spaces. Running it locally requires Git, conda, FFmpeg, Python 3.10, and downloaded pretrained weights, so it is a better fit for users comfortable with an AI project setup than for those seeking a turnkey editor.

Portrait animation can produce visual artifacts and can be misused to create deepfakes. The maintainers disclaim legal responsibility for generated results; consent and intended use warrant care.

Key features

Multiple motion inputs: Driving video, audio, text, or generation can supply motion for a still portrait. That range makes the project useful for varied animation workflows, while its source-video plus driving-video mode also supports portrait video editing.

Facial controls: Stitching and eye- and lip-retargeting modules give users more control over the result than a single-pass animation workflow.

Animal animation: Fine-tuned models can animate cats, dogs, and pandas from human driving motion. This broadens the creative use case, but the mode has narrower hardware and platform support than human animation.

Reported speed: The project reports generation at 12.8 ms on an RTX 4090 GPU with PyTorch. That is a useful performance reference for that hardware, not a guarantee for other systems.

Privacy option: Motion templates can speed inference and protect privacy, which may suit workflows where sharing a driving video is undesirable.

Pricing

LivePortrait is free and open source under the MIT License, with no paid plan described. The trade-off is that free access does not remove setup work: users need the required software dependencies and pretrained weights, and local performance depends on their hardware.

The commercial-use condition matters. InsightFace detection models included in the project are for non-commercial research; commercial users must remove and replace those models to comply with their license. The project’s MIT license does not eliminate that separate requirement.

Platforms

Human mode supports Linux, Windows, and macOS on Apple Silicon. Intel macOS is not tested in the stated instructions, so users on those machines should not assume the same compatibility.

Animal mode is tested only on Linux and Windows with an NVIDIA GPU. It is unsupported on macOS because its X-Pose dependency does not support that platform. The hosted Hugging Face Spaces demo provides a web route to try LivePortrait without setting up the local repository, while the project itself offers a self-hosted workflow.

Who it's for

LivePortrait suits developers, technically confident creators, and researchers who want a free local portrait-animation system with multiple motion inputs and retargeting controls. Its animal models are a particular draw for Linux or Windows users with an NVIDIA GPU. It is less suitable for someone who needs a polished, low-setup commercial service or a supported animal workflow on macOS.

Pros and cons

Pro: Free, MIT-licensed code and local inference make it possible to build a portrait-animation workflow without a paid plan.

Pro: Video, audio, text, and generated motion, plus stitching and eye/lip retargeting, offer meaningful control over facial animation.

Pro: Fine-tuned cat, dog, and panda models extend the project beyond human portraits.

Con: Local setup requires several tools, Python 3.10, and separate pretrained weights, creating friction for nontechnical users.

Con: Animal animation excludes macOS and requires an NVIDIA GPU on its supported Linux and Windows configurations.

Con: Commercial users must replace the included InsightFace detection models, adding a licensing and implementation step.

Con: The maintainers warn of visual artifacts and deepfake misuse, so outputs may need scrutiny and responsible handling.

Alternatives

For a hosted option with a stated free monthly video allowance, consider HeyGen: its free plan includes up to three videos per month, each no longer than one minute, across web, Android, iOS, and API. Pika is another freemium web, iOS, and API option, though its free plan has zero monthly credits and uses packs instead.

Hedra may suit users who prefer monthly credit plans and commercial use: Basic costs $20 a month for 2,000 credits, while Pro costs $50 a month for 7,200 credits. For a text-to-video, image-to-video, or reference-based world-generation model, consider Vidu.

Animate Photo AI offers a web-based free tier with 50 credits, which may be preferable for a browser-first workflow. AnimateDiff is a free, self-hosted alternative released for academic use. For one-time credit packs or a monthly paid plan, AnimatePhoto is another web option. Animate Old Photos may fit users seeking a paid service with a 7-day cloud history and one-time credit packs.

For more options, browse AI Photo Animation Tools and Photo Animation Software.

Verdict

Choose LivePortrait if you want free, locally runnable portrait animation with flexible motion inputs and facial retargeting, and you can manage its dependencies and licensing conditions. Look elsewhere if you need a simpler hosted workflow, dependable Intel Mac support, or animal animation on macOS.