Reeldo AIReeldo AI

OmniHuman 2.0 generator

OmniHuman 2.0 AI Talking Avatar Generator

You already have the face and the track — turn them into a 720p lip-sync video with expression and full-body motion, up to five minutes per run.

  • Up to 5 min lip sync
  • 720p talking avatar
  • Speech & singing

Step 1

Portrait photo

One clear face, eyes and mouth visible. Neutral expression works best. JPEG, PNG, or WebP up to 30 MB.

Step 2

Performance audio

MP3 or WAV, 4 seconds to 5 minutes. Trim silence at the ends to save credits.

Tip: test a 10-second clip before a long take. Create narration first

≈ 20 credits

Script → AI Voiceover or mic recording → OmniHuman 2.0 → trim tail → publish or upscale.

OmniHuman 2.0 is a ByteDance research model made available through Reeldo. Reeldo is not affiliated with ByteDance.

How it works

You bring the performance audio. Reeldo handles lip sync, expression, and motion.

  1. 01

    Upload a clear portrait

    One face, front-facing, neutral expression works best. Frame the shot how you want the video — vertical photos become vertical clips.

  2. 02

    Add MP3 or WAV

    Speech, song, or dialogue — 4 to 300 seconds. Trim long silences at the ends so you do not pay for dead air.

  3. 03

    Optional direction line

    Short notes like calm presenter or small hand gestures. Up to 300 characters.

  4. 04

    Generate & download

    Credits are shown before you run. Preview, download MP4, trim any silent tail in your editor.

  • OmniHuman 2.0 is an AI talking avatar model: one portrait plus one performance audio becomes a 720p lip-sync video with expression and body motion.
  • Creators who already have narration or vocals and need a virtual presenter, UGC spokesperson, singing avatar, or animated character — without filming.
  • Multi-reference scene rebuilds, motion copies, or @tag workflows — use Reference Studio. Need MP3 first? Use AI Voiceover.

Create narration (AI Voiceover)Back to generator

AI talking avatar use cases

Same two inputs — portrait plus audio. Pick the job that matches your pain point.

  • Virtual presenter talking avatar lip sync on laptop

    01

    Virtual presenter lip-sync

    You need on-camera explainers without booking a shoot or reshooting every script change.

    Outcome

    Turn a headshot and narration into course intros, updates, and welcome videos in 720p.

  • AI singing avatar performance preview

    02

    AI singing avatar

    You want a character to perform a vocal track — not just nod along to a beat.

    Outcome

    Upload a singing take; the portrait lip-syncs and moves with the rhythm of your audio.

  • UGC spokesperson talking video on phone frame

    03

    UGC spokesperson ads

    You need spokesperson-style hooks at scale without new talent every week.

    Outcome

    Batch short ad reads from one approved photo plus fresh voice recordings.

  • Illustrated character talking avatar lip sync

    04

    Photo-to-talking character

    Your mascot or OC only exists as a still — you need it to perform dialogue or song.

    Outcome

    Animate illustrations, fan art, or brand characters with one performance audio file.

Photo + audio

What you get from each run

Three outputs every time — lip sync video, length that follows your audio, and optional direction.

01

Lip sync that follows your audio

Mouth, eyes, and shoulders track speech or singing.

More detail

Mouth, eyes, and shoulders track speech or song — built for explainers, hooks, and vocal takes.

Audio length

4s
5 min

Max 300s per run · billed per second

02

Up to five minutes, one take

Output length matches your track — 4 to 300 seconds per run.

More detail

Carry a full narration from one portrait. Split very long scripts if motion softens on extreme length.

Direction prompt controls for avatar motion

03

Optional direction line

Up to 300 characters for mood, gestures, or camera feel.

More detail

Steer mood, gestures, or camera feel in up to 300 characters — alongside your audio.

OmniHuman 2.0 vs 1.5

This page runs 2.0 only. Use the table when tutorials or search results still mention 1.5.

TopicOmniHuman 2.0OmniHuman 1.5
On ReeldoYes — generator at top of this pageNot available on Reeldo today
Typical max lengthUp to 300 seconds (5 minutes) per audio fileOften capped near 60 seconds per job
Output resolution720p MP4720p or 1080p depending on provider
Direction promptOptional — up to 300 charactersNot on standard 1.5 lip-sync routes
Best forLong narration, lessons, full vocal takesShort ads, hooks, 1080p social cuts

Need only a 15–30s 1080p hook? OmniHuman 1.5 can fit — on Reeldo, validate your portrait with a 10-second proof on 2.0 first.

Credits & length

Billing follows your audio length (rounded up), at about 5 credits per second.

  • 10s

    10-second proof ≈ 50 credits

  • 60s

    One-minute read ≈ 300 credits

  • 5 min

    Full five-minute take ≈ 1500 credits

After you download

  • Trim silent tail and add B-roll
  • Regenerate only the audio if the script changed
  • Need new scenes? Use Image to Video for B-roll
  • Upscale final master if delivery requires more pixels

If the result feels off, fix one variable at a time — usually photo, audio, or length.

  • Clip slightly longer than audio

    Trim a half-second at the end in CapCut or Premiere — common on talking-avatar models.

  • Background shimmers

    Re-upload with a plain wall or soft blur behind the subject.

  • Motion softens on long takes

    Split into two audio files under three minutes, generate separately, and edit together.

  • Mouth feels loose

    Use dry vocals without music under the voice; re-export a clean MP3.

OmniHuman 2.0 FAQ

Length, credits, portraits, watermarks, and how 2.0 differs from 1.5 or Reference Studio.

What is OmniHuman 2.0?

How does Reeldo run OmniHuman 2.0?

OmniHuman 2.0 vs 1.5 — which should I use?

How long can a video be?

What resolution do I get?

Does this page generate speech?

How are credits calculated?

Does the MP4 include a watermark?

What photo works best?

Can I use singing vocals?

Does it support multiple people?

What should I write in the direction line?

Which languages work in the direction line?

OmniHuman 2.0 vs Reference Studio?

Why is my file longer than the audio?

Can I use output in ads?

Why is there a consent checkbox?

What if generation fails?

Is this an official ByteDance product page?