Prepare the inputs
Choose one well-lit portrait and a short driver clip with a similar camera angle. Crop distractions and keep the face large enough to read clearly.
Practical tutorial
A clear workflow for how to use live portrait ai
This guide shows how to prepare a source image, choose a driving video, run a LivePortrait animation, and improve the result without guessing at every setting.
Prerequisites
A good input does more for consistency than aggressive adjustments. Check these essentials before starting your first run.
- Source imageA forward-facing portrait with visible eyes and an unobstructed mouth gives LivePortrait clearer facial landmarks.
- Driver clipA short driving video with one visible face makes motion transfer easier to control.
- FramingMatching the subject's framing and head angle reduces stretching around the jaw and hairline.
- ExpectationThe final result depends on the source, driver, and selected settings rather than on a prompt alone.
Numbered steps
The basic workflow has three stages: prepare clean inputs, transfer motion, then inspect and refine the output.
Run the animation
Upload the source image and driving video, then start with the default motion settings. Let the first pass reveal whether the inputs are compatible.
Review and refine
Watch the eyes, mouth, jaw, hair, and background across the full clip. Replace the driver or tighten the crop before making complex adjustments.
Common errors and fixes
Most disappointing outputs come from input mismatch, not from a lack of advanced controls. Use this checklist before changing several variables at once.
Required
Use a single, clearly visible face in the source image.
Multiple faces can confuse the subject selection.
Required
Choose a driver with steady framing and moderate facial movement.
Fast turns and cuts can create unstable transitions.
Required
Keep the source and driver at similar head angles.
Large pose differences often distort the cheeks or jaw.
Required
Use even lighting without heavy shadows across the eyes.
Strong contrast can make expressions look uneven.
Optional
Trim long pauses, camera cuts, or unrelated movement from the driver.
A shorter useful segment is easier to evaluate.
Optional
Save the original image and clip before cropping or converting them.
This makes later comparisons easier.
Useful paths to explore
Once the basic workflow is clear, these related guides explain where to run the model and how its input-output options differ.
-
1
liveportrait online
Compare browser-based ways to test a LivePortrait workflow without building a local interface.
-
2
liveportrait ai
See how the AI model transfers facial motion from a driving performance to a still portrait.
-
3
liveportrait video to video
Understand when a video-driven workflow is useful and what changes between source and output clips.
Advanced tips
Different audiences need different levels of control. Start with the simplest panel that matches your goal, then move to a more configurable workflow only when needed.
Start with a clean default pass
Use one portrait and one short driver, keep the default settings, and judge the complete clip before making changes. This gives you a baseline and prevents several small edits from hiding the real cause of an artifact.
- Prioritize a large, unobstructed face.
- Use moderate expressions before dramatic gestures.
- Compare the first and last frames for drift.
Optimize for a repeatable sequence
Prepare a consistent source-image style and record driver clips with the same camera distance. Repeatable framing makes a series of animated portraits look more coherent and makes failed takes easier to diagnose.
- Keep lighting and camera height consistent.
- Name source and driver files clearly.
- Review short sections before exporting a full sequence.
Tune one variable at a time
When you need more control, change one setting or one input at a time and retain the previous result. Separate input problems from rendering problems by testing the same driver against a different portrait.
- Keep a baseline output for comparison.
- Test motion changes on a short clip first.
- Inspect edges, hair, eyes, and mouth separately.
How the workflow evolved
LivePortrait moved from a research-style motion-transfer workflow toward simpler hosted experiences, but the core logic remains the same: source, driver, inference, review.
-
Motion transfer enters the spotlight
Research demonstrations made it easier to see how a driving facial performance could animate a still portrait while preserving much of the source identity.
-
Community workflows expand
Users began adapting the model through notebooks, demos, node graphs, and hosted interfaces, creating more ways to experiment with the same basic inputs.
-
Input quality becomes the practical focus
Tutorials and examples increasingly emphasized portrait framing, driver selection, lighting, and artifact review rather than relying on a single magic setting.
-
A repeatable three-stage method
The most dependable approach remains preparation, a conservative first pass, and targeted refinement after watching the complete animation.
Compare the first pass
A before-and-after comparison is most useful when the framing stays stable. Look for natural eye motion, mouth shape, jaw edges, and background consistency.
- Source
- Animated result
Drag the divider to inspect facial motion and edge stability.
Put the workflow into practice
Create a first LivePortrait animation
Use the preparation checklist, begin with a conservative driver, and review the complete result before refining it. A clear baseline makes every later improvement easier to measure.
- Start with one source and one driver
- Review the complete output
- Refine only after identifying the artifact
Tutorial FAQ
These answers address the questions people commonly ask when learning a LivePortrait workflow from the beginning.
You need a suitable source portrait and a driving video that shows the facial motion you want to transfer. A clear face, steady framing, and compatible head angles usually matter more than complicated settings.
Choose a clip with one visible face, steady lighting, and the expression or head movement you want. Avoid rapid cuts, severe profile turns, heavy occlusion, and movements that are much larger than the source portrait can support.
Distortion often comes from a poor source crop, mismatched head angles, fast driver motion, or shadows across important facial features. Test a cleaner portrait or calmer driver before changing several controls at once.
The simplest workflow starts with a still portrait and uses a video as the motion driver. If your goal is video-to-video transformation, use a workflow designed for that input pattern and check how it handles identity, framing, and temporal consistency.
Use moderate expressions, keep the face well framed, and choose a driver whose pose resembles the source. Watch the full clip for eye, mouth, hair, and background artifacts instead of judging only a single attractive frame.