OmniHuman 2.0: Audio-Driven AI Video Generator

Upload one image and an audio track, add a prompt if you want, and this lip sync video generator returns a 720p avatar video up to 5 minutes long. Start with 20 free credits.

Try OmniHuman 2.0 free See credits and plans

OmniHuman 2.0 Studio

One image, your audio, and the performance you have in mind.

Example preview

Paris Street Fashion Ad

Prompt

Keep original composition, character positions, outdoor café scene, table layout…

Input image for Paris Street Fashion Ad — Input image

Input audio

What is OmniHuman 2.0?

An audio-driven digital human model from ByteDance that turns one image and one audio track into a speaking, singing or acting video.

OmniHuman 2.0 is an audio-driven digital human generation model from ByteDance, offered through BytePlus. Provide one image, one audio track and an optional prompt to generate a 720p digital human video up to 5 minutes long.

Compared with OmniHuman 1.5, it extends the maximum length per generation from 30 to 300 seconds and, BytePlus says, improves long-duration character consistency, motion expressiveness and visual quality. It also understands what the audio says, so the character can act on it.

Most general-purpose video models start from text or visuals and use audio only as a reference. OmniHuman 2.0 works the other way round: the audio shapes the performance, including lip movements, intonation and emotion.

That makes it an AI avatar video generator and AI lip sync video generator for full-length music videos, marketing presentations, educational explainers, virtual IP, motion comics and product introductions. Some people search for it as OmniHuman V2 or OmniHuman 2.0 AI.

OmniHuman 2.0 Key Features

Six things this audio-driven model does that a plain lip-sync tool does not.

Audio Comprehension

The character understands what the audio says and acts on it. If the audio mentions taking a blue pill, the character can show the pill. The meaning of the audio, not only its sound, drives the performance.

Prompt Control

Direct the scene like a video model. Chain several actions, such as walking forward, talking with hands on hips, then turning to watch an explosion. Prompts also control the camera and background.

Emotional Performance

OmniHuman 2.0 recognizes the emotion in your audio and lines and performs it, with the strongest results in dialogue scenes. OmniHuman lip sync follows the original audio: lip movements, intonation and emotion.

Up to 5 Minutes in One Run

One generation can reach 300 seconds, enough for a full song, a course segment or a podcast clip, with less stitching. BytePlus says 2.0 improves long-duration character consistency compared with 1.5.

Multi-Person Scenes

Choose which person in the frame speaks. The other characters keep natural movements instead of standing still, so a conversation or group shot works from one image.

Animals, Pixel Art and Cartoons

Your OmniHuman 2.0 avatar can be a cat, a dog, a pixel-style chick or an anime character, so the OmniHuman AI avatar is not limited to photos of people.

What Can You Create With OmniHuman 2.0?

Explore four creative directions for OmniHuman 2.0, from film and music to imaginative personal content.

Film, Television, and Short Video Production

Create character dialogue for short dramas, creative advertising, documentary narration and brand stories with expressive acting and varied camera work.

Fantasy Vlog Production

Combine a realistic selfie-video style with imaginative scenes, from time-travel interviews to animal-led vlogs and fictional documentaries.

AI MV and Music Content

Turn music into performances with rhythm-driven actions, emotional expression and dynamic camera movement for music videos and musical scenes.

UGC Creative Content

Make personal creative videos with cartoon characters, pets, pixel art and virtual avatars. Use prompts to add actions and enrich the scene.

OmniHuman 2.0 Video Examples

Explore OmniHuman 2.0 video examples across live-action scenes, fantasy characters, music and more. Open each example to see its input image, audio and original prompt.

Live-Action Storytelling and Commercial Content

2 examples

Paris Street Fashion Ad

Inputs

Input audio

Original prompt

Keep original composition, character positions, outdoor café scene, table layout, coffee cups, pastries, and warm late-afternoon lighting unchanged.Realistic cinematic fashion advertisement video, based on the uploaded reference image and audio. Keep the same beautiful long-haired European supermodel, avant-garde coat, ankle boots, Paris street setting, golden hour lighting and French stone architecture.The model walks briskly towards the camera on a straight Paris sidewalk, with natural confident body movement. As she walks, she slowly and naturally removes her sunglasses with one hand, then looks directly at the camera. She speaks the fashion advertisement line while walking, with accurate lip-sync, natural facial expressions and supermodel aura.The camera follows her walking pace with a smooth forward tracking shot, slowly pushing closer. Shallow depth of field, golden hour street lighting, realistic skin texture, film grain, high realism, cinematic fashion commercial style.Do not slow down the walking motion. Do not change the model's appearance, outfit, hairstyle or original Paris street atmosphere.Negative prompt: face distortion, stiff gestures, robotic lip sync, camera shake, slow motion, changed character appearance, changed café layout.

Output video

Natural Conversation at an Outdoor Café

Inputs

Input audio

Original prompt

Keep original composition, character positions, outdoor café scene, table layout, coffee cups, pastries, and warm late-afternoon lighting unchanged.

The long-haired woman in the center, brown leather jacket and orange-red scarf, speaks naturally. She slightly leans forward, rests both hands near the table, and makes small casual hand gestures while talking. Her eyes shift naturally between the two friends across the table.

The man on the left listens attentively with a gentle smile and subtle head nodding. The woman on the right listens quietly, looks at the speaking woman, and nods slightly.

Add realistic details: soft breeze moves hair and scarf, slight steam from coffee cups, faint background pedestrians, natural eye blinking, realistic lip sync.

Cinematic camera: very slow subtle push-in from the center, gentle left-right micro-wobble like a handheld long take, soft focus breathing, shallow depth of field, light film grain.

All movements slow, natural and restrained. No exaggerated acting.

Negative prompt: face distortion, stiff gestures, robotic lip sync, camera shake, slow motion, changed character appearance, changed café layout.

Output video

Fantasy, Animals, and Complex Scenes

2 examples

Dark Fantasy Wizard Narrative

Inputs

Input audio

Original prompt

Cinematic dark-fantasy medium shot, slow forward push-in. Keep the bearded sorcerer in the flowing black robe, his staff, the black warhorse beside him, the sword planted in the ground, the cliff edge, heavy sunset clouds and distant stone towers exactly as in the reference image. Wind lifts his robe and hair; the horse shifts its weight and tosses its mane; faint magical embers drift from the staff tip. He turns his head toward the camera and speaks the line with a low, commanding tone, accurate lip-sync, steady confident gaze and subtle jaw tension. Warm rim light from the setting sun on his face. Photorealistic, Arri Alexa, anamorphic lens, volumetric light, subtle film grain. Negative prompt: face distortion, robotic lip sync, changed outfit or scene, camera shake, extra limbs.

Output video

3D Raccoon Character Performance

Inputs

Input audio

Original prompt

Whimsical cinematic 3D-animation style, gentle forward dolly. Keep the anthropomorphic raccoon, its purple cape and scarf, the wooden staff, the mossy stone path and sun-dappled dense forest unchanged. The raccoon looks toward the camera, small ears twitch, whiskers move, one paw lifts the staff slightly as it speaks the line with expressive cartoon lip-sync, bright curious eyes and a mischievous half-smile. Light breeze moves leaves and floating pollen; soft god-rays filter through the canopy. Pixar-like rendering, shallow depth of field, warm forest color grade. Negative prompt: uncanny human mouth, stiff face, changed costume, flicker, camera shake.

Output video

Show all 10 examplesShow fewer examples

Cinematic Characters and Dynamic Performances

2 examples

Emotional Singing in a City at Night

Inputs

Input audio

Original prompt

Emotional cinematic medium shot, slow orbit around the singer. Keep the young woman, her light long-sleeve dress, the handheld microphone and the blue-to-orange gradient sky with distant city lights unchanged. She sings the lyrics with expressive, accurate singing lip-sync, brows lifting on high notes, chest breathing, subtle head tilts and closed-then-opening eyes conveying emotion. Long hair and dress flow in the wind. Golden-to-blue dusk lighting on her face, cinematic concert color grade, shallow depth of field, subtle film grain. Negative prompt: robotic mouth, stiff neck, off-beat lip sync, changed appearance, camera jitter.

Output video

Dynamic Running Performance in the Countryside

Inputs

Input audio

Original prompt

Bright cinematic tracking shot, camera moves backward following her run. Keep the young light-blonde girl, her white shirt and dark shorts, the sunlit dirt country road, white wildflowers, low stone wall, wooden houses and green hills unchanged. She runs joyfully toward the camera, kicking up light dust, hair bouncing, then laughs and calls out the line mid-stride with natural breathless lip-sync and bright sparkling eyes. Warm golden sunlight, lens flare, vibrant summer grade, shallow depth of field. Negative prompt: slow motion, stiff running, face distortion, changed outfit, camera shake.

Output video

Anime Characters and IP Content

2 examples

Anime Character Presentation in Snow

Inputs

Input audio

Original prompt

Fresh anime-style medium shot, gentle forward drift with falling snow. Keep the Japanese-style girl with black long hair, hooded dark coat and blue-brown plaid scarf, snow on her shoulders, and the snowy riverside winter countryside unchanged. She smiles, breath fogging in the cold air, snowflakes drifting, and speaks the line with soft anime lip-sync, warm eyes and a gentle head tilt. Crisp bright winter anime lighting, soft bloom, drifting snow particles. Negative prompt: realistic mouth, stiff face, changed design, flicker, camera shake.

Output video

Anime Boy Presentation on a Blue Background A

Inputs

Input audio

Original prompt

Clean anime-style medium shot, subtle push-in on a flat blue background. Keep the young male character with black short hair, white shirt over a white tee, headphones around his neck and a beige shoulder bag unchanged. His serious expression eases slightly as he looks at the camera and speaks the line with crisp anime lip-sync, a single confident nod and natural blink. Bright even anime lighting, clean cel-shaded style. Negative prompt: realistic mouth, stiff body, changed design, background clutter, flicker.

Output video

Music and Podcasts

2 examples

Anime character singing

Inputs

Input audio

Original prompt

Anime-style cinematic medium shot, gentle forward drift. Keep the white-haired blue-eyed boy in the white shirt, the bright blue sky, white clouds and drifting petals unchanged. He turns his head toward the camera, hair and petals moving in the wind, and speaks the line with soft anime lip-sync, gentle blinking and a subtle emotional smile. Bright dreamy anime lighting, soft bloom, floating petals, shallow depth of field. Negative prompt: realistic mouth, stiff face, flicker, changed character design, camera shake.

Output video

Podcast Studio Presentation

Inputs

Input audio

Original prompt

Professional cinematic medium shot, slow push-in. Keep the man in the white tee and olive-green jacket wearing studio monitor headphones, the broadcast microphone in front of him, and the warm-lit recording studio with plants and wooden decor unchanged. He leans slightly toward the mic and speaks the line with a calm, professional tone, a focused nod, natural blinking and accurate lip-sync. Warm studio lighting, shallow depth of field, cinematic grade, film grain. Negative prompt: robotic lip sync, stiff posture, changed studio, camera shake, face distortion.

Output video

Who is OmniHuman 2.0 Built For?

OmniHuman 2.0 puts a talking or singing character on screen without a camera crew.

For marketing and e-commerce teams

Make presentations and product introductions with a consistent presenter. Change the audio and the prompt for each product or ad variant, keep the same face, and test a new version the same day.

For film, short-drama and video creators

Shoot character dialogue, documentary narration and brand stories with prompts that control action and camera. Short-video creators can post pets, cartoons and virtual avatars without filming themselves.

For musicians and music-video makers

Upload a full song and one image of the artist or a character. OmniHuman 2.0 matches rhythm and emotion, and prompts add camera moves for a stronger music-video feel.

For educators and podcasters

Turn a course segment, explainer or podcast episode into a video with a presenter. One generation reaches 5 minutes, so long audio needs less stitching.

What Changed From OmniHuman 1.5 to OmniHuman 2.0?

A quick look at the two latest OmniHuman versions, using only what each source publishes.

	OmniHuman 1.5	OmniHuman 2.0
Input	One image and audio	One image, audio, optional prompt
Max length per generation	30 seconds	300 seconds
Resolution	Not published	720p
Prompt, camera and character movement	Limited, per BytePlus	Prompts control actions, camera and background
Other changes BytePlus lists	—	Long-duration character consistency, motion expressiveness, visual quality
Technical write-up	Research paper (arXiv 2508.19209)	Not published separately; BytePlus links to the 1.5 paper
Where to run it	BytePlus, fal.ai, Replicate, ElevenLabs, Kie.ai	BytePlus API, and this site

OmniHuman 2.0 raises the maximum length per generation from 30 to 300 seconds. According to BytePlus, it also improves long-duration character consistency, motion expressiveness and visual quality, and adds prompt control over action, camera and background.

OmniHuman 1.5 is ahead on openness: it has a public research paper, and more providers offer it. We have not run a head-to-head test, and no public benchmark compares the two.

Pick OmniHuman 2.0 for long clips and prompt control. Pick 1.5 for its published method or a provider you already use. Read the full OmniHuman 2.0 vs OmniHuman 1.5 comparison.

Compare OmniHuman 2.0 and 1.5 examples

How To Make a Video With OmniHuman 2.0

Three steps from one image and one audio file to a downloadable OmniHuman 2.0 video.

Upload one image and one audio file. Use an image under 30 MB and a WAV, MP3 or M4A file of 4 to 300 seconds. OmniHuman 2.0 image to video starts here: you can make an image lip sync to any recording.
Write a prompt (optional). Describe an action, an object or a camera move in Chinese, English, Japanese, Korean, Mexican Spanish or Indonesian. Skip this step to let the audio lead. See the OmniHuman 2.0 prompt guide for examples.
Generate and download. Press Generate, then download the finished 720p clip, up to 5 minutes long. The cost is shown before you confirm. For tips, read how to use OmniHuman 2.0 in our guide.

Read the OmniHuman prompt guide

How Much Does OmniHuman 2.0 Cost?

You pay in credits: 2 credits for each second of 720p video, and 20 free credits to start.

Credit packsMonthly plans

A fresh credit balance each month for ongoing video creation.

Monthly plan

Starter

Start with short clips and explore your ideas.

$29.90/ month

170 credits

About 8 short videos

About 85 seconds: 8 × 10-second or 2 × 30-second clips

Based on 10-second clips at 720p. Longer videos use more credits.

720p, up to 5 minutes per video
WAV, MP3 or M4A audio upload
Prompt control for actions, camera, background
Multi-person, pet, cartoon and pixel characters
Auto-renews monthly; cancel any time
Failed videos refunded; credits never expire

Recommended

Monthly plan

Pro

More room for regular content and creative experiments.

$49.90/ month

500 credits

About 25 short videos

About 250 seconds: 25 × 10-second or 8 × 30-second clips

Based on 10-second clips at 720p. Longer videos use more credits.

720p, up to 5 minutes per video
WAV, MP3 or M4A audio upload
Prompt control for actions, camera, background
Multi-person, pet, cartoon and pixel characters
Auto-renews monthly; cancel any time
Failed videos refunded; credits never expire

Monthly plan

Enterprise

A larger credit balance for more video production.

$199.90/ month

2,500 credits

About 125 short videos

About 1,250 seconds: 125 × 10-second or 41 × 30-second clips

Based on 10-second clips at 720p. Longer videos use more credits.

720p, up to 5 minutes per video
WAV, MP3 or M4A audio upload
Prompt control for actions, camera, background
Multi-person, pet, cartoon and pixel characters
Auto-renews monthly; cancel any time
Failed videos refunded; credits never expire

A 10-second OmniHuman 2.0 clip costs 20 credits, so the 20 free credits cover one 10-second video. A 60-second OmniHuman 2.0 clip costs 120 credits. The shortest audio you can use is 4 seconds, which costs 8 credits. Check the estimate on the generator before you press Generate. See the full OmniHuman 2.0 pricing page.

OmniHuman 2.0 FAQ

The most common questions about the OmniHuman 2.0 model, answered.

What is OmniHuman 2.0?

OmniHuman 2.0 is an audio-driven digital human generation model from ByteDance. Give it one image, one audio track and an optional prompt, and it generates a 720p video up to 5 minutes long.

The character's lip movements, intonation and emotions follow the audio, and the model can act on what the audio says. Compared with OmniHuman 1.5, it extends the maximum length from 30 to 300 seconds and, BytePlus says, improves long-duration character consistency, motion expressiveness and visual quality.

It suits full-length music videos, marketing presentations, educational explainers, audio-content visualization, virtual IP, motion comics and product introductions. This site is an independent tool built on the model.

How long can an OmniHuman 2.0 video be?

Up to 5 minutes. Audio can run from 4 to 300 seconds, and OmniHuman 2.0 generates up to 300 seconds in a single run, so full songs and course segments need less stitching. Output is 720p.

OmniHuman 1.5 was limited to 30 seconds per generation. Longer runs matter most for character consistency, which BytePlus says improved in 2.0. Test with a short clip first, then run the full length once the result looks right.

What image and audio does OmniHuman 2.0 accept?

OmniHuman 2.0 takes one image and one audio file. Images can be JPEG, PNG or WebP, up to 30 MB, from 300 to 6000 pixels, with a width-to-height ratio from 0.4 to 2.5.

Audio must be WAV, MP3 or M4A, 4 to 300 seconds long and under 15 MB. Songs, speeches, dialogue and narration all work. Use a clear image where the subject is easy to see.

Can OmniHuman 2.0 understand what the audio says?

Yes. OmniHuman 2.0 reads the meaning of the audio and can act on it. If the audio mentions an object, the character can show that object.

Emotion follows the audio too, with the strongest results in dialogue scenes. Add a prompt for more control over actions, camera and background.

Can OmniHuman 2.0 make a singing video?

Yes. Upload a song as the audio and OmniHuman 2.0 performs it, with rhythm-driven actions, emotion matching and natural pauses across music genres.

Add a prompt such as a slow camera zoom if you want the shot to move. Songs let the character use more camera moves and body movement, which gives the video a stronger music-video feel.

Does OmniHuman 2.0 support multiple people in one video?

Yes. OmniHuman 2.0 supports multi-person scenes. You can specify which person in the frame speaks, and the other characters keep natural movements.

Use it for a short dialogue, an interview or a group shot from a single image. Keep every person clearly visible in the image.

Can OmniHuman 2.0 animate pets and cartoon characters?

Yes. OmniHuman 2.0 supports more styles and subjects than people alone, including cats, dogs, cartoon and anime characters, pixel-style chicks and penguins, and virtual avatars.

Add a prompt to supplement elements in the picture, such as an object or a setting, while the audio drives the performance.

How is OmniHuman 2.0 different from a general-purpose video model?

OmniHuman 2.0 is audio-driven: the audio directly shapes the character's performance. General-purpose video models usually start from text or visuals and use audio only as a reference.

OmniHuman 2.0 generates up to 300 seconds in one run, while general-purpose models suit short clips better. They are the better choice for shot variation and scene generation; OmniHuman 2.0 is built for digital human performance, presentations, singing, music videos and education.

Try OmniHuman 2.0 With 20 Free Credits

Upload one image and one audio clip to make your first OmniHuman 2.0 video.

Try OmniHuman 2.0 free See credits and plans

Videos come out at 720p. Only upload images and audio you have the right to use, and do not use anyone's face or voice without their permission.