1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
import requests
import base64
# Use this function to convert an image file from the filesystem to base64
def image_file_to_base64(image_path):
with open(image_path, 'rb') as f:
image_data = f.read()
return base64.b64encode(image_data).decode('utf-8')
# Use this function to fetch an image from a URL and convert it to base64
def image_url_to_base64(image_url):
response = requests.get(image_url)
image_data = response.content
return base64.b64encode(image_data).decode('utf-8')
api_key = "YOUR_API_KEY"
url = "https://api.segmind.com/v1/hallo"
# Request payload
data = {
"input_image": image_url_to_base64("https://segmind-sd-models.s3.amazonaws.com/display_images/hallo/hallo-input.png"), # Or use image_file_to_base64("IMAGE_PATH")
"input_audio": "https://segmind-sd-models.s3.amazonaws.com/display_images/hallo/hallo-audio-2.mp3",
"pose_weight": 1.4,
"face_weight": 1,
"lip_weight": 1.2,
"face_expand_ratio": 1.2,
"base64": False
}
headers = {'x-api-key': api_key}
response = requests.post(url, json=data, headers=headers)
print(response.content) # The response is the generated imageAttributes
Input image of a talking-head.
Input audio file. Avoid special symbol in the filename as it may cause ffmpeg erros.
Weight of the Pose in output
min : 0,
max : 10
Weight of the Face in output
min : 0,
max : 10
Weight of the Lip to apply
min : 0,
max : 10
Face Expand Ratio
min : 0,
max : 10
Base64 encoding of the output image.
To keep track of your credit usage, you can inspect the response headers of each API call. The x-remaining-credits property will indicate the number of remaining credits in your account. Ensure you monitor this value to avoid any disruptions in your API usage.
Hallo
Hallo is novel technique for generating animated portraits that seamlessly blend audio with facial movements. Creating lifelike portrait animations presents a unique challenge. It's not just about lip syncing – the animation needs to capture the full spectrum of human expression, from subtle eyebrow raises to head tilts, while maintaining visual consistency and realism. Existing methods often struggle to achieve this, resulting in animations that appear uncanny or unnatural. Hallo tackles this challenge with a hierarchical audio-driven visual synthesis module. This module acts like a translator, interpreting audio features (speech) and translating them into corresponding visual cues for the lips, facial expressions, and head pose.
Under the hood of Hallo
Imagine two spotlights focusing on different aspects – the audio and the visuals. The cross-attention mechanism ensures these spotlights work together, pinpointing how specific audio elements correspond to specific facial movements. The animation process leverages the power of diffusion models, which excel at generating high-quality, realistic images and videos. Maintaining temporal coherence across the animation sequence is crucial. The method incorporates this by ensuring smooth transitions between frames. A "ReferenceNet" component acts as a guide, ensuring the generated animations align with the original portrait's unique features. The method offers control over expression and pose diversity, allowing creators to tailor the animations to their specific vision.
Use cases
Hallo significantly improves the quality of generated animations, creating more natural and realistic talking portraits. Additionally, the lip synchronization and overall motion diversity are vastly enhanced. This opens doors for captivating new forms of storytelling and content creation. With the ability to animate portraits and imbue them with speech, applications range from personalized avatars to interactive learning experiences.
Popular Models
