7 Real-World Applications

What People Are Actually Building With SadTalker

Specific use cases based on what SadTalker's feature set actually supports — with the settings, tradeoffs, and limitations for each.

01. E-Learning 02. Reviving Old or Historical Photos 03. Animating AI-Generated Art 04. Marketing 05. Redubbing 06. Accessibility 07. Business Presentations

Choose a workflow below, prepare a clear portrait and narration, and inspect a short clip before producing the final video. Useful controls include: --still mode, --preprocess full, the GFPGAN enhancer, and the Stable Diffusion WebUI and video-lip-sync extensions the community has built around it.

01

E-Learning & Virtual Instructors

A single instructor photo turned into a narrating avatar for course modules, onboarding videos, or training slides — recorded once, then reused with new audio for every new lesson without re-filming.

What you would actually use

--still mode

Keeps the presenter composed, avoiding distractions from content.

512px face model

Clearer detail on training platforms viewed full-screen.

Multilingual reuse

Same source image with different audio tracks in different languages — SadTalker syncs lip movement to whatever audio it's given.

Where it falls short

Long-form lessons (20+ minutes) are impractical to render this way — this fits short modules and micro-lessons better than full lecture recordings.

Suggested settings

Preprocess

crop

Still

Enhancer

Optional

02

Reviving Old or Historical Photos

Taking a single archival photo — a historical figure, an old family photo, a painting or illustrated portrait — and animating it to "speak" a piece of narration, quote, or scripted dialogue. One of the most common community use cases.

What you would actually use

--preprocess full

For half-body or full-body images rather than tight face crops.

GFPGAN enhancer

Compare restoration OFF/ON on older or soft scans; it may generate plausible detail rather than recover the original face exactly.

--still

Start with restrained head movement. Pose style selects a motion pattern; it is not a motion-strength slider.

Where it falls short

Very low-resolution or heavily damaged source photos produce visibly distorted results — the enhancer helps but doesn't fully compensate for a poor source image.

Suggested settings

Preprocess

full

Still

Enhancer

Compare ON/OFF

Prepare an old photo for animation

Use a scan with one visible face. Crop away album borders and other people; keep the eyes, nose and mouth unobstructed.
Use full preprocessing to retain the photograph’s composition, or crop for a close talking-head view. Start with still mode ON and enhancement OFF to inspect the original likeness.
Pair the image with a short narration. Listen for clipping, background music and long pauses before rendering.
Check the mouth, eyes and pasted face boundary. Compare GFPGAN separately if facial detail needs restoration.

These are starting settings for an old-photo workflow. For measured settings comparisons, see full vs crop vs resize and GFPGAN ON vs OFF.

Common problems to check

Several faces or a profile view can make face selection unreliable. Isolate one clear face first.
Scratches across the mouth or eyes can carry into the animation. Compare the output with the source at the same size.
Restoration may change wrinkles, teeth or eye detail. Keep enhancement OFF when preserving likeness matters more than sharpening.

Narration from the website

Download narration

Starting command

Run from your SadTalker installation folder after saving the input image and audio there. Replace old-photo.png with your scan.

python inference.py --source_image ./old-photo.png --driven_audio ./sadtalker-male-narration-audio.wav --result_dir ./results/old-photo --preprocess full --size 512 --still --batch_size 2 --pose_style 0 --expression_scale 1.0

Adjust the command · Use ComfyUI

03

Animating AI-Generated Art & Characters

Taking a Stable Diffusion-generated portrait (an original character, a stylized avatar, concept art) and giving it a voice — a workflow the community formalized with a dedicated SD WebUI extension for SadTalker.

What you would actually use

SD WebUI extension

Generate the face and animate it in the same Automatic1111 pipeline.

Standard crop

Straightforward preprocessing for face-on character portraits.

Where it falls short

Highly stylized or non-photorealistic art (extreme cartoon proportions, non-human anatomy) wasn't the training distribution SadTalker was built on — results are less reliable outside realistic-to-semi-stylized human faces.

Suggested settings

Preprocess

crop

Still

On for a steady presenter

Enhancer

Optional

Avatar input and animated result

This existing website example pairs a digital-avatar portrait with male narration. The playable output is 11.24 seconds at 25 fps, with a 1024 × 1024 export.

Input portrait

Animated output with narration. Export dimensions do not indicate the base renderer size.

In the inspected frame at 4 seconds, facial texture and the background appear softer than in the input, and the mouth shape changes. Inspect the moving clip around speech and pauses before choosing the final version.

This example uses a human-shaped face. Exaggerated cartoon features, covered mouths and non-human anatomy need separate testing. Compare enhancement OFF if restoration alters the character’s design.

Narration from the website

Download narration

Example command for this workflow

Run from your SadTalker installation folder after saving the input image and audio there. The example uses crop, 512, still ON and GFPGAN. It is a recipe to run, not a fresh timing measurement.

python inference.py --source_image ./sadtalker-male-avatar-source.png --driven_audio ./sadtalker-male-narration-audio.wav --result_dir ./results/avatar --preprocess crop --size 512 --still --enhancer gfpgan --batch_size 2 --pose_style 0 --expression_scale 1.0

Download avatar input

Adjust the command · Use ComfyUI

04

Marketing & Localized Ad Campaigns

A single brand spokesperson image reused across multiple regional ad variants — same face, different audio per market or language — instead of re-filming a presenter for every locale.

What you would actually use

512px model

For customer-facing output where visual polish matters more than render speed.

Batch generation

Multiple audio files for localizing one script into several languages.

Where it falls short

Not a substitute for a full production shoot when a campaign needs full-body motion, product interaction, or complex scene composition — suited to talking-head-format ads and social clips.

Suggested settings

Preprocess

crop or full

Still

Depends on brand tone

Enhancer

Compare ON/OFF

05

Redubbing & Lip-Sync for Existing Video

Changing the audio track of an existing talking video — dubbing, correcting a line, or translating spoken content — while keeping the lip movement in sync with the new audio, not the original.

What you would actually use

Video-Lip-Sync extension

The community-maintained SadTalker-Video-Lip-Sync extension, built specifically for re-syncing lips on video input rather than animating a still image.

Where it falls short

This is a distinct workflow from the core single-image tool. If your input is already a video (not a photo), you want the video-lip-sync extension specifically, not the base image-to-video pipeline.

Suggested settings

Preprocess

Use lip-sync extension

Still

—

Enhancer

—

06

Accessibility & Visual Communication Aids

Pairing spoken audio with a visible, expressive face for audiences who benefit from a visual anchor alongside audio — reinforcing spoken announcements, presentations, or instructional content with a synchronized face.

What you would actually use

--still mode

A calm, non-distracting presenter.

Clear, well-lit source photos

Facial clarity matters more here than in entertainment — the visual is doing communicative work, not just decorative work.

Where it falls short

SadTalker generates lip and facial motion from audio — it does not generate or interpret sign language, so it's a complement to (not a replacement for) dedicated sign-language content.

Suggested settings

Preprocess

crop

Still

Enhancer

Compare ON/OFF

07

Business Presentations & Corporate Communication

A consistent virtual presenter face for internal updates, product announcements, or investor communications — recorded once as a source image, reused across many scripted updates.

What you would actually use

--still + GFPGAN

A polished, professional look suitable for external-facing communication.

Where it falls short

For anything requiring gestures, props, or a full-body presence (e.g., a product demo where hands matter), this format is the wrong tool — it's built for head-and-shoulders talking presenters.

Suggested settings

Preprocess

crop

Still

Enhancer

Compare ON/OFF

Quick Reference

Choosing the Right Settings by Use Case

A summary of recommended configuration for each workflow.

Use case	Preprocess	Still mode	Enhancer
E-learning / instructor	`crop`	On	Optional
Old photo revival	`full`	On	Compare ON/OFF
AI art / character	`crop`	On for a steady presenter	Optional
Marketing / ads	`crop or full`	Depends on brand tone	Compare ON/OFF
Video redub	(use lip-sync extension)	—	—
Accessibility content	`crop`	On	Compare ON/OFF
Business presentation	`crop`	On	Compare ON/OFF