Turn your own photo, an original character or an AI-generated fictional portrait, plus one audio track, into a video of that character speaking or singing. Up to 5 minutes in a single take.
Add a portrait and a voice track to continue.
Free preview · limited generations per hour · 720p · audio 4 s – 5 min · Own or consented photos only, no impersonation of real people.
See how it works Powered by OmniHuman 2.0, available via BytePlus · Independent site
Get early access updates
0:00 / 0:00
0:00 / 0:00
0:00 / 0:00
0:00 / 0:00
0:00 / 0:00
How it works
No rigging, no keyframes, no separate singing mode. Bring your own face or an original character and a sound — the model does the performing.
portrait.jpgA selfie, a portrait of someone who has given you consent, an original clay or drawn character, or an AI-generated fictional portrait. A plain background and a single clear face work best.
Speech, a song, or a dialogue — it's all the same input. There's no separate singing mode to switch on.
OmniHuman 2.0 renders a video of that character performing the audio — face, head and body — in one continuous take of up to 5 minutes.
TIP — a plain background and a single clear face give the best results. Only use photos you own or have consent for.
Features
OmniHuman 2.0 hears tone, emotion and words — and turns all three into a believable on-camera performance.
5:00
The previous version capped out at 60 seconds. Now you get up to 5 minutes rendered continuously — no stitching.
The model listens for emotion and delivery. Laughs, whispers and pauses show up on the face, not just in the soundtrack.
Mouth shapes follow each syllable — including fast lines and long held notes in songs.
It generates face, head and body motion together, so the result reads as a performance rather than a pasted-on mouth.
Stitched clips — visible cuts
One take — continuous
Because nothing is stitched together, there are no sudden shifts in the face, the lighting or the background mid-video.
photo
clay
drawnWorks with your own selfies and consented portraits, AI-generated fictional people, clay figures, drawn characters and brand mascots. Never real or famous people without their consent.
Same input. No singing mode.
One model handles narration, singing and conversation. Just bring the audio — the model figures out the rest.
↓ cost
vs. previous version · per BytePlus
According to BytePlus, OmniHuman 2.0 is cheaper to run than the previous version — which means more minutes for your budget.
Spec sheet
Everything OmniHuman 2.0 needs, and everything it gives back.
Before / after
Typical lip-sync tools animate a mouth. OmniHuman 2.0 animates a person.
Mouth onlyUse cases
From five-minute stories to fifteen-second shorts. Portraits by Nano Banana 2.1, voices by ElevenLabs, video by OmniHuman 2.
Turn a script and an instructor photo into long-form lessons — up to 5 minutes per take. This sample is a calm, outdoor gardening lesson.
Hover or scroll to preview · generated with OmniHuman 2
Pricing
Two simple monthly plans, shown in full before you pay. Checkout is handled securely by Creem.io.
$9.99/month
Billed monthly
35 credits per month
= 35 seconds of video every month.
$29.99/month
Billed monthly
120 credits per month
= 120 seconds of video (2 min) every month.
Prices in USD. Taxes calculated at checkout by Creem.io. Credits renew monthly and do not roll over.
Responsible use: all plans are for your own or explicitly consented photos and for original or fictional characters. Impersonating real people or celebrities, deepfakes and voice cloning are prohibited and prompts naming real people are blocked. See the Acceptable Use Policy.
FAQ
Can't find what you need? Everything about inputs, length and availability is here.
OmniHuman 2.0 is ByteDance's audio-driven human video model, released via BytePlus on October 15, 2026. You give it one photo you own (or have consent to use) and one audio track, and it generates a video of that character performing the audio.
Just two things: a single photo and a single audio track. The photo must be your own, of someone who gave you explicit consent, or of an original or AI-generated fictional character. The audio can be speech, a song or a dialogue. You confirm this with a required consent checkbox before every upload.
No. Singing uses exactly the same input as speech — upload the song and the model handles it, including fast lyrics and long held notes.
The OmniHuman 2.0 model supports up to 5 minutes in one continuous take (the previous version was capped at 60 seconds), and the video isn't stitched from shorter pieces, so the face, lighting and background stay consistent. On OmniHuman2.net each video is 4 to 120 seconds long.
No. Plain lip sync only moves the mouth. OmniHuman 2.0 generates the face, head and body performance, and it reacts to tone and emotion — so laughs, whispers and pauses show on the face.
Yes. Besides your own selfies and consented portraits, it works with original clay and drawn characters, mascots and AI-generated fictional portraits.
Use a photo with a plain background and a single, clearly visible face. Clean audio without heavy background noise also helps.
No. You may only use your own photo and voice, or those of a person who has given you clear, explicit consent. Impersonating celebrities, public figures or other real people, deepfakes, face swaps and voice cloning are prohibited, prompts that name or imitate real people are blocked, and text-to-voice only offers stock synthetic voices. See our Acceptable Use Policy.
Yes. Create a free account, then pick a plan. Starter is $9.99/month and includes 35 credits (35 seconds of video per month). Pro is $29.99/month and includes 120 credits (2 minutes of video per month). Both are billed monthly and you can cancel anytime from your dashboard.
1 credit = 1 second of generated video, rounded up to the next full second, with a 4-second minimum and a 120-second maximum per video. Credits are reserved when a video starts and refunded automatically if it fails. Credits renew with each monthly payment and unused credits do not roll over. A generated voice from text costs 1 credit per 100 characters.
No. OmniHuman2.net is an independent product built on third-party AI models. It is not affiliated with, endorsed by, or sponsored by ByteDance, BytePlus, fal.ai or ElevenLabs. All trademarks belong to their respective owners. NSFW, explicit, illegal content, deepfakes and impersonation of real people are prohibited — see our Acceptable Use Policy.
One photo · one track · up to 5:00
One photo, one track, up to five minutes of continuous performance.
Get early access updates
Photo + audio → video
Up to 5 min in one take
