Skip to main content
AI video model
ByteDance
OmniHuman

OmniHuman 1.5

The current OmniHuman 1.5 core path turns one image of a person, pet, or animated subject plus one audio clip into a driven video. Signed upload metadata verifies audio duration, which directly determines pre-task authorization.

Availability
Limited access to core capabilities
Input
1 image + 1 audio clip
Output
720P / 1080P
Pricing
27 / audio second
30-second overview
  • A clear input contract of one image plus one audio clip
  • 720P / 1080P output and fast mode
  • Signed audio duration directly enters preauthorization
  • The auxiliary mask workflow is unavailable and mask fields are hard-rejected

Popular searches: How do I use OmniHuman 1.5? · How much does OmniHuman 1.5 cost? · How long can OmniHuman 1.5 audio be?

Examples

Cinematic scene

Mukbang ASMR Vlog CAMERA/LOOK: Handheld DV 16mm camcorder POV, held by CHASE (never visible on screen), occasionally propped for stability. Shaky, imperfect framing, delayed focus pulls, clumsy zooms, occasional face-cut-off selfie framing. Soft/blurry tape quality, faint noise, bloomy highlights, flickering auto-exposure, muted contrast, natural skin tones. STYLE: Mukbang x ASMR-lite. Quiet, minimal dialogue, whispered/soft-spoken lines only. Slow cuts, close framing, stillness. Ambient sound focus: chopsticks, sizzling, steam, soft chewing, quiet sighs. CHASE: Korean idol, 20s, long straight black hair, dewy skin, coral lips, big eyes, slim-curvy build. Cozy oversized hoodie/loungewear, minimal jewelry. Soft, relaxed energy; occasional glances/smiles at lens. SETTING: Dorm room at night, warm lamp light, small food table, camera angled to capture her + food, low ambient noise. STORYBOARD: (2s, propped, medium) Sits on floor cushion, adjusts camera, soft wave. "Hey guys... doing a quiet mukbang tonight." (2s, close insert) Steam off noodle bowl, chopsticks lift a bite — sizzle sound only. (2s, handheld, face track) Slow bite, eyes close, soft exhale. "Mmm... this is so good." (1.5s, macro, shallow DOF) Chopsticks lift another piece, steam/texture close-up — clink sound only. (2s, medium) Warm smile at lens, head tilt. "I need to order this again sometime." (2s, close/propped) Quiet sip, sets cup down, content sigh — ambient only. (1.5s, tight punch-in) Chin on hand, gazing at food. "This is my favorite part of the day." (2s, slow fade-out) Reaches to end recording, soft smile, hand blocks lens as it cuts. "Goodnight, everyone."

Cinematic scene

A cinematic 15-second luxury jewelry ad featuring a confident young woman in a sleek black silk off-shoulder dress standing in a softly lit, high-end marble and glass penthouse at golden hour. The camera starts with a close-up of her hands slowly adjusting a diamond tennis bracelet, light catching every facet. It then pans upward to reveal layered heavy diamond necklace, matching statement earrings, and multiple diamond rings sparkling subtly as she moves. Soft, elegant orchestral ambient music builds. The lighting is warm but controlled, with sharp reflections of city skyline bokeh in the background. She turns slightly toward the camera with a calm, powerful expression, not speaking. Slow-motion shot as she brushes her hair back, diamonds catching intense light flares. Final 3 seconds: extreme close-up of the jewelry set with brand space left clean for logo and tagline like “Timeless Brilliance. Infinite Glow.” Fade out on sparkling diamonds.

Cinematic scene

Create a 15 second, 16:9 realistic viral social media video that looks like authentic low resolution smartphone footage, not a polished film or AI generated scene. 0.0–8.6 seconds: A wide, elevated handheld smartphone shot from the top of a steep sandy hill overlooking a quiet natural pond. A long, bright blue plastic tarp has been laid down the entire slope like a homemade water slide, continuing all the way to the edge of the water. A shirtless, heavyset adult man wearing black knee length shorts lies flat on his stomach at the top of the wet tarp, facing downhill. Someone just outside the frame gives him a small push. He rapidly slides belly first down the slippery blue plastic, becoming smaller as he travels farther away. The phone camera remains near the top of the hill and follows him with a slightly shaky downward tilt and a subtle digital zoom. Show pale sand around the tarp, a parked white vehicle and two tiny people near the shoreline, calm grey-blue water, green marsh patches and thin trees in the distance. The man reaches the narrow final section of the tarp and shoots directly onto the surface of the pond without sinking. He unexpectedly skims across the water at high speed while still lying forward, producing two strong symmetrical trails of white spray and a long widening V shaped wake behind him. The moment should initially feel impressive, funny and believable, with realistic water physics, ripples and imperfect phone-camera focus. 8.6–9.3 seconds: Just as the man reaches the distant side of the pond, suddenly insert a crude, frightening jump-scare overlay directly in the center of the screen: a grotesque horror clown with messy orange red hair, pale cracked skin, an unnaturally wide mouth, sharp teeth and intensely glowing red eyes lunges rapidly toward the camera. The clown should appear abruptly and fill most of the frame, like an old internet prank video, with rough compositing and compressed visual quality rather than cinematic horror effects. 9.3–10.2 seconds: Hard cut to a separate handheld phone recording inside a small pink bedroom. A young woman with long straight dark hair, large clear eyeglasses and a light-colored floral dress sits casually on a bed, holding a blue smartphone close to her face as though she has just watched the waterslide video. Behind her is a pink decorated wall with hanging medals, small framed pictures, butterfly stickers and decorative writing; a pale curtain hangs on the right, and the room has a simple corrugated metal ceiling. Butterfly-print pillows are visible on the bed. The jump scare frightens her instantly. Her eyes widen, mouth opens, shoulders jerk upward, both arms fly into the air and the phone slips violently from her hands. She recoils and collapses backward against the pillows in an uncontrolled, genuinely startled reaction. Use strong natural motion blur, slight camera shake and awkward body movement. 10.2–15.0 seconds: Freeze the video on the funniest reaction frame: the woman leaning backward on the bed, mouth open, head tilted, one arm raised and the phone caught mid air beside her. Hold this exact frozen frame until the video ends, preserving the blurry compressed look of a reposted meme clip. Visual direction: Authentic amateur smartphone recording, ordinary daylight, slightly washed out colors, low bitrate, visible compression, inconsistent autofocus, mild motion blur, imperfect framing and natural handheld shake. Compose the outdoor action through the center of the horizontal 16:9 frame while showing additional pond, shoreline, sand and trees on both sides. Keep all people anatomically consistent before the jump scare. No cinematic grading, no dramatic camera orbit, no slow motion, no smooth drone movement, no subtitles, no captions, no logos, no watermark, no dialogue and no background music.

Cinematic scene

Main Subject : Young Korean woman, early 20s, oversized knit sweater over shorts, hair loose and wind-tossed, relaxed joyful expression, firelight glow on her face. Location : Quiet Korean beach during dusk, small bonfire setup. Driftwood logs as seats, dim string lights, waves faintly visible in background, sand dunes. No crowds, no commercial beach setup. Visual Style : Warm nostalgic realistic documentary tone, cozy firelit atmosphere, soft dusky lighting. Camera Style : 1990s VHS camcorder, handheld with warm firelight flicker affecting exposure, soft grain, muffled ambient mic. No stabilization. Timeline (15 sec, each slot = 2 compressed beats): 00:00–00:03 → She sits by the fire, then pokes it gently with a stick. 00:03–00:06 → She looks up at the camera saying "불 진짜 따뜻하다" ("The fire is really warm"), smiling. 00:06–00:09 → She hugs her knees, then watches sparks rise into the dark sky. 00:09–00:12 → She laughs at something off-camera, then says "너도 이리 와서 앉아" ("Come sit here too"). 00:12–00:15 → She smiles warmly at the camera, then looks back at the fire as it fades to black. Audio : Crackling fire, waves in distance, wind, faint laughter. Her dialogue as noted above. No music. Goal : A cozy, warm beach evening memory intimate, joyful, real.

Fashion portrait

Cinematic scene

1962 lost television horror anthology episode, authentic Victorian mansion ballroom at night, elegant aristocrats dancing beneath crystal chandeliers, candlelight, live orchestra, luxurious gothic atmosphere, Kodak 16mm film, analog television broadcast, subtle film grain, soft studio lighting, practical effects, psychological vampire horror, authentic 1960s cinematography, natural dialogue, approximately 15 seconds, 10 cinematic cuts Shot 1 (0.0s-1.5s) Wide shot. A grand ballroom filled with elegant guests dancing. A charming gentleman approaches a beautiful woman. Dialogue: Man: "May I have this dance?" Cut Shot 2 (1.5s-3.0s) Medium shot. They begin dancing together. Dialogue: Woman: "You're surprisingly cold." Man: "I've been cold for a very long time." Cut Shot 3 (3.0s-4.5s) Close-up of the man's smile. His teeth appear perfectly normal. Cut Shot 4 (4.5s-6.0s) Medium close-up while they continue dancing. Dialogue: Woman: "Where are you from?" Man: "A place people never leave." Cut Shot 5 (6.0s-7.5s) Close-up of the man's mouth as he smiles again. His canine teeth have become slightly longer than before, almost unnoticeable. Cut Shot 6 (7.5s-9.0s) The woman stares at him, confused. Dialogue: Woman: "Your teeth..." Man: "Do they frighten you?" Cut Shot 7 (9.0s-10.8s) Extreme close-up. His smile grows wider. His fangs visibly lengthen another few centimeters while she watches in horror. Cut Shot 8 (10.8s-12.3s) Wide shot. The orchestra stops. Every guest slowly turns toward them. Each guest smiles as their own vampire fangs begin growing. Cut Shot 9 (12.3s-13.8s) The woman backs away. Dialogue: Woman: "What are all of you?" Man: "Hungry." Cut Shot 10 (13.8s-15.0s) Close-up. His fangs extend to their full length as he leans toward her with a calm smile. The guests surround her. Film flicker, analog distortion, abrupt cut to black.

At a glance

Is this model right for you?

Images can use any aspect ratio but must still pass true-format and signed-dimension checks. Audio must use a supported format, stay within 10 MB, and be under 60 seconds. The optional prompt follows the stricter 300-character limit rather than a wider conflicting description.

What it does best
A core audio-driven video workflow for a single-subject image.
Who it is for
A good fit for presenter clips, virtual-character dialogue, and talking videos built from pet or animated-subject images.
Why use it on Kyeo AI
Kyeo verifies the image/audio binding before reserving credits, preauthorizes from trusted audio seconds, and clearly disables mask capabilities that are not integrated.

Key facts

Quickly evaluate whether this model fits your business use case.

Image formats
JPEG / PNG / WebP
Audio duration
Under 60 seconds
Prompt
Optional; up to 300 characters
Not currently available
Subject-mask workflow

Other names

The same model may have different names in official documentation, community discussions, and API integrations. This section brings them together for you.

OmniHuman V1.5
OmniHuman 1.5 talking image
OmniHuman 1.5 audio-driven video

Common questions

These practical questions focus on the task, input conditions, and result requirements you should confirm before choosing.

How do I use OmniHuman 1.5?
How much does OmniHuman 1.5 cost?
How long can OmniHuman 1.5 audio be?
What image formats does OmniHuman 1.5 support?
What does OmniHuman 1.5 fast mode do?

Decision guide

Still deciding? Start with these key decision points.

1
Choose OmniHuman for a still image

Use this model when the source is a static person or character image that needs to be driven by audio.

2
Compare lip sync when you already have video

If the original motion and camera need to be preserved, compare Video Lip Sync.

3
Test with short audio first

Use a short recommended audio sample to check subject recognition and lip movement before extending the script.

Popular comparisons

Compare common alternatives on the same task to clarify differences in inputs, controls, and cost.

How do the input and pricing of OmniHuman 1.5 and Kling AI Avatar differ?
When should I choose OmniHuman 1.5 or a video lip-sync model?

Model comparison table

Not sure which model to choose? Compare the key differences between this model and its alternatives at a glance.

Core input
OmniHuman 1.5
1 image + 1 audio clip
Kling AI Avatar Standard
1 image + 1 audio clip
Video Lip Sync
1 video + 1 audio clip

Real-world experience

Practical guidance based on public sources, current on-site limits, and representative tasks.

Limited access is not a complete auxiliary suite

The core generation path is callable, but multi-subject selection that depends on mask detection is not part of the Kyeo runtime contract.

Per-audio-second preauthorization is verifiable

Trusted duration is known before task creation, avoiding fixed-price guesses or user-supplied duration estimates.

Capabilities

Image-and-audio driving

Create a talking-subject video from one image and one audio clip.

Two output resolutions

Choose 720P or 1080P.

Fast mode

Offers a faster-processing option with a possible quality tradeoff.

Use cases

Virtual presenter

Use a clear, front-facing character image and clean narration to make an explainer clip.

Animated-character dialogue

Generate a speaking character from an animated-subject image and dialogue audio.

Talking-pet clip

Pair a pet photo with short audio to experiment with anthropomorphic expression.

Prompt tips

Keep the subject clear

Choose an image with a prominent subject, minimal occlusion, and a recognizable face or head.

Use clean speech audio

Reduce background music and noise to make rhythm and lip alignment easier.

Keep the prompt concise

Stay within 300 characters and focus on expression, movement range, and camera style.

Why choose it

Simple and explicit image-plus-audio input.
Audio duration can be priced precisely before task creation.
Supports multiple subject types and any image aspect ratio.
Unavailable mask functionality has a clear boundary.

What to know first

Mask-based or specific multi-subject workflows are not currently available.
Only one image and one audio clip can be submitted.
Audio must be under 60 seconds.
The prompt uses a strict 300-character limit.

Adjustable settings

Quickly evaluate whether this model fits your business use case.

Output resolution
Optional
Parameter type: Select
Default: 1080P
720P
1080P
Fast mode
Optional
Parameter type: Toggle
Default: off
Off
On
Random seed
Optional
Parameter type: Integer
Default: -1 (random)

Credit cost

27 credits per audio second

Kyeo calculates preauthorization from verified audio duration at 27 credits per second and rounds the total up. Audio must be under 60 seconds, so a valid request is preauthorized for less than 1,620 credits. Final settlement uses actual consumption after completion.

Credit-saving tip

Before generating images or videos in bulk, run an A/B test with the same input on this model and similar alternatives. Confirm the result first to avoid wasting credits.

Frequently asked questions

Related models

Still deciding? Take a quick look at these similar alternatives.

Kling AI Avatar Standard

Workflow: The UI requires exactly 1 JPEG or PNG image and 1 MPEG, WAV, X-WAV, AAC, MP4, or OGG audio file. The image may be up to 10 MB; audio may be up to 100 MB and 5 minutes. The charge uses the signed audio duration at 8 credits per second and rounds up. A missing, invalid, or over-5-minute duration cannot be submitted.

AI video model
8 credits / audio second

Infinitalk

Infinitalk uses 1 image, 1 audio file no longer than 15 seconds, and a required prompt to generate a speaking-portrait video on Kyeo. It offers 480p or 720p plus optional `seed`; cost depends on audio duration and resolution rather than a fixed request price.

AI video model
480p: 3 credits/sec; 720p: 12 credits/sec

Volcengine Video Lip Sync

Video Lip Sync rebuilds mouth movement in one existing video to match a target vocal track. Lite mode targets front-facing single-person video, while Basic mode handles more complex single-person scenes and can enable scene detection.

AI video model
8 credits per audio second

Sources

This page is based on vendor materials, model-family documentation, and Kyeo AI integration settings.

Source note

Explore the currently available OmniHuman 1.5 workflow: one image plus one audio clip, 720P or 1080P, audio under 60 seconds, per-second pricing, and limited-access boundaries. Kyeo calculates preauthorization from verified audio duration at 27 credits per second and rounds the total up. Audio must be under 60 seconds, so a valid request is preauthorized for less than 1,620 credits. Final settlement uses actual consumption after completion.

Last updated: 2026-08-28
OmniHuman V1.5 current interface notes
Integration

Used to check the input modes, visible controls, media limits, and availability of the version currently connected to this site. The page and workbench show the actual options.