veo-3-1/image-to-video
Create realistic motion visuals with Veo 3.1's sleek AI video conversion.
A portrait only fixes who is on screen. The recording decides how they perform. OmniHuman 2.0 reads speech, sung lines, and dialogue for tone, emotion, and wording, then builds mouth shapes and small gestures that follow that track.
Short avatar clips used to stop around a minute, which meant cutting and joining pieces. Here the same take can run on OmniHuman 2.0 for up to five minutes, from a four-second minimum through a 300-second maximum.
| Advantage | What you get |
|---|---|
| Track-led acting | OmniHuman 2.0 puts tone and word stress in the face and hands, not only in a repeating mouth cycle. |
| One unbroken take | Audio from 4 to 300 seconds stays a single clip. You do not join minute-long fragments. |
| Face from the still | OmniHuman 2.0 keeps identity from the uploaded portrait. Timing comes from the file you attach. |
| Light direction | OmniHuman 2.0 can take a short note for calmer hands or a brighter expression. Skip it and the audio still leads. |
You do not type a script for OmniHuman 2.0 to speak. The words, pauses, and energy already live in the audio file. A prompt only nudges how OmniHuman 2.0 moves the body while that file plays.
On RunComfy you can try a take in the model UI, then repeat the same image, audio, and prompt through the HTTP API.
A busy background, a tiny face, or a track full of noise will show up in the result. Start with a simple portrait and a clean voice take, then lengthen the audio once the performance looks right.
image, audio, and optional prompt through the HTTP API.These are the only controls on this OmniHuman 2.0 page.
| Parameter | Required | Type | Default | Range / Options | Description |
|---|---|---|---|---|---|
image* | Yes (*) | String | Example portrait | Image URL | Portrait that sets the character. Use a clear, front-facing face. |
audio* | Yes (*) | String | Example audio | 4-300 seconds | Recording that sets speech, timing, and performance. |
prompt | No | String | Example line | Text | Optional note for expression, movement, or style. |
There is no resolution picker, seed, or duration slider. Clip length follows the audio you upload.
OmniHuman 2.0 bills the processed audio length at $0.09 per second. The clock rounds up to the next whole second. The shortest OmniHuman 2.0 charge is 4 seconds and the longest is 300 seconds.
| Audio length | Billed seconds | Price |
|---|---|---|
| 4s | 4 | $0.36 |
| 10s | 10 | $0.90 |
| 30s | 30 | $2.70 |
| 60s | 60 | $5.40 |
| 300s | 300 | $27.00 |
Based on publicly available information. Length limits below describe the inputs those tools publish, not a scored ranking.
| Model | Inputs | Length | Standout |
|---|---|---|---|
| OmniHuman 2.0 | Still portrait and audio, optional prompt | 4-300 seconds in one shot | Acting follows tone and wording, not only lip shapes |
| Omnihuman 1.5 | Still portrait and audio, optional prompt and seed | Under 35 seconds | Earlier ByteDance avatar pass with a seed for repeatable takes |
| Video lip-sync | An existing face video plus new audio | Depends on the source clip | Keeps a finished shot and mainly retimes the mouth |
Pick this model when you start from a still and need the voice to carry a long, unbroken performance. Use a lip-sync tool, not OmniHuman 2.0, when the picture is already a video and only the mouth should change.
Create realistic motion visuals with Veo 3.1's sleek AI video conversion.
Generates natural speech and audio from text, reference audio, or an image
Generates up to 4-minute songs with vocals and lyrics from text tags
Reshape a source clip from a text prompt with native audio.
Pro-tier reference-driven 3-15s video generation from $0.19 per second.
Transforms static visuals into expressive motion clips with sync sound
OmniHuman 2.0 turns a portrait and an audio file into a talking-avatar video. The image sets who appears, and the recording sets timing, tone, and mouth movement. It fits virtual hosts, lessons, and social clips where speech or music should drive the performance.
OmniHuman 2.0 accepts audio from 4 to 300 seconds and keeps that length as one continuous shot. Omnihuman 1.5 on RunComfy is limited to audio under 35 seconds and includes a seed. This version does not expose a seed, and it follows wording and emotion rather than mouth shape alone.
OmniHuman 2.0 requires a portrait image and an audio file. A prompt is optional and only steers expression, gesture, or style. There is no resolution, seed, or duration control on this page, because clip length follows the audio.
OmniHuman 2.0 is meant to read speech, sung lines, and dialogue, including tone and emotion, then perform them. A clean recording still behaves more predictably than a noisy one. The prompt can name the kind of performance, such as a lecture or a chorus, without replacing the audio.
Audio for OmniHuman 2.0 must be between 4 and 300 seconds. Billing rounds up to the next whole second, with a 4-second minimum and a 300-second maximum. If an upload is rejected, check the current RunComfy parameter panel for file type and size limits.
Keep the OmniHuman 2.0 prompt to one clear note about expression, gesture, or style, and do not contradict the audio. You can leave the prompt empty when the recording should lead on its own. Test a new portrait on a short clip before rendering several minutes.
Prototype OmniHuman 2.0 in the RunComfy Playground Web UI, then call the same model through the RunComfy API with the same image, audio, and optional prompt. Each generation spends usd credits. The API uses the same parameters and does not add a separate uptime guarantee.
OmniHuman 2.0 costs $0.09 per billed second of audio. Duration rounds up to the next whole second, from 4 seconds through 300 seconds. A 10-second clip is $0.90, and a 60-second clip is $5.40.
RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.