OmniHuman 2.0
Talking avatar from a photo and audio: lips and expression follow speech or a song. 4 to 300 seconds, 720p.
Good for: voice-overs, host avatars, song clips and explainer videos
How to get a better result
"Replace the background with a beach", "make it a watercolor".
The AI keeps everything else better: the face, the pose, the clothes.
Sharp, not blurry, so the person or object is clearly visible.
Pass this model in the request omnihuman-2.0
1. Create a task POST /api/v1/jobs/createTask
curl -X POST https://mixmedia.tech/api/v1/jobs/createTask \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: order-123" \
-d '{
"model": "omnihuman-2.0",
"callBackUrl": "https://your-app.example/callback",
"input": {
"prompt": "A friendly guide talks about the stars, soft smile, steady camera",
"image_urls": [
"https://example.com/photo.jpg"
],
"resolution": "720p"
}
}'import time, requests
API = "https://mixmedia.tech/api/v1"
HEADERS = {"Authorization": "Bearer YOUR_API_KEY"}
task = requests.post(f"{API}/jobs/createTask", headers=HEADERS, json={
"model": "omnihuman-2.0",
"callBackUrl": "https://your-app.example/callback",
"input": {
"prompt": "A friendly guide talks about the stars, soft smile, steady camera",
"image_urls": [
"https://example.com/photo.jpg"
],
"resolution": "720p"
}
}).json()
task_id = task["data"]["taskId"]
while True:
info = requests.get(f"{API}/jobs/recordInfo", headers=HEADERS, params={"taskId": task_id}).json()["data"]
if info["state"] in ("success", "fail"):
break
time.sleep(4)
print(info["state"], info.get("resultJson") or info.get("failMsg"))const API = "https://mixmedia.tech/api/v1";
const headers = { Authorization: "Bearer YOUR_API_KEY", "Content-Type": "application/json" };
const { data } = await fetch(`${API}/jobs/createTask`, {
method: "POST", headers,
body: JSON.stringify({
"model": "omnihuman-2.0",
"callBackUrl": "https://your-app.example/callback",
"input": {
"prompt": "A friendly guide talks about the stars, soft smile, steady camera",
"image_urls": [
"https://example.com/photo.jpg"
],
"resolution": "720p"
}
}),
}).then((r) => r.json());
let info;
do {
await new Promise((r) => setTimeout(r, 4000));
info = (await fetch(`${API}/jobs/recordInfo?taskId=${data.taskId}`, { headers }).then((r) => r.json())).data;
} while (!["success", "fail"].includes(info.state));
console.log(info.state, info.resultJson ?? info.failMsg);Response: data.taskId. callBackUrl and Idempotency-Key are optional.
2. Status and result GET /api/v1/jobs/recordInfo
curl "https://mixmedia.tech/api/v1/jobs/recordInfo?taskId=TASK_ID" \
-H "Authorization: Bearer YOUR_API_KEY"
data.state: waiting, queuing, generating, success, fail. On success data.resultJson is a JSON string with resultUrls.
3. Upload a file POST /api/v1/files/upload
curl -X POST https://mixmedia.tech/api/v1/files/upload \
-H "Authorization: Bearer YOUR_API_KEY" -F "file=@photo.jpg"
The response has data.url. Pass that URL in the file field. Any public https URL works too.
About the model
OmniHuman 2.0 by ByteDance animates one photo to your audio — speech, a song or a track. Facial expression, lip sync and light head motion follow the sound; the clip length matches the audio.
You need one character photo (face large, front-facing) and audio from 4 to 300 seconds (MP3/WAV). An optional note up to 300 characters can set pose, camera and mood. Resolution 720p. You pay per second of your audio.


