OmniHuman 1.5 Digital Human Video
Generate a talking-head digital human video from one portrait + one voice track. Great for explainers, education, virtual hosts, and customer support (async).
Video
Direct URL to an MP3/WAV file (max 35s). Must be publicly accessible.
Subject check
Verify a human subject exists in the image (slower start).
Auto mask
Automatically generate a mask.
Fast mode
Faster generation with potentially lower quality.
Tip: use a clear, front-facing portrait (bust/half-body) and clean audio; remember to download and save results promptly.
Covers talking heads, emotional performance, singing rhythm, multi-character dialogues, and more.
Loading demos…
Use cases
What can OmniHuman 1.5 do?
Turn “avatar + voice” into deliverable video content, ideal for fast production and multi-scenario reuse.
Product selling / explanations
Use one consistent avatar to output multiple short selling videos.
Online courses / knowledge content
Keep the lecturer avatar consistent and swap audio across languages.
Customer support / how-to guides
Turn FAQs and instructions into visual explainers.
Virtual anchors / IP avatars
News readouts, livestream scripts, and event hosting.
Run your first generation in 3 minutes
Prepare assets and follow the steps to generate.
- 01/ 04
Prepare a reference portrait
Front-facing, clear, unobstructed; bust/half-body recommended.
- 02/ 04
Add audio
Upload audio or paste a link (MP3/WAV).
- 03/ 04
(Optional) refine with text
Use short prompts to control style/camera/motion.
- 04/ 04
Generate and download
Preview and download after generation; save promptly.
OmniHuman 1.5 best practices
Focus on “clear portrait + clean audio.”
Practice 1: clear, front-facing portrait
Avoid side faces, occlusions, and extreme lighting.
Practice 2: clean audio at a moderate pace
Reduce noise and reverb for more stable lip sync.
Practice 3: short prompts for style control
Example: “natural light, half-body, slight nod.”
FAQ
FAQ
Get started
Start creating with OmniHuman 1.5
Use the generator above: write a prompt (and optionally upload inputs), tune params, then generate.