API
If you're looking for an API, you can choose from your desired programming language.
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
34
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/insta-depth"
# Request payload
data = {
"prompt": "Photo of a woman wearing a Superman costume",
"negative_prompt": "lowquality, badquality, sketches",
"face_image": image_url_to_base64("https://segmind-sd-models.s3.amazonaws.com/display_images/insta-depth-ip.png"), # Or use image_file_to_base64("IMAGE_PATH")
"pose_image": image_url_to_base64("https://segmind-sd-models.s3.amazonaws.com/display_images/insta-dept-pose.png"), # Or use image_file_to_base64("IMAGE_PATH")
"num_inference_steps": 10,
"guidance_scale": 3,
"seed": 354849415,
"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
Prompt to render
Prompts to exclude, eg. 'bad anatomy, bad hands, missing fingers'
Face Image.
Pose Image.
Number of denoising steps.
min : 4,
max : 100
Scale for classifier-free guidance
min : 1,
max : 15
Seed for image generation.
min : -1,
max : 999999999999999
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.
Insta Depth
Insta Depth generates new images that very closely resemble a specific person and also allowing for different poses and angles. It achieves this using a single input image of the person and a text description of the desired variations, along with a reference image. The key aspect of Insta Depth involves transferring the composition of the person’s image into different poses based on the face image of the person. This ensures that the generated image maintains the unique identity of the person while allowing for variations in pose. This model is an add-on and improvement to InstantID model.
Key Components of Insta Depth
Insta Depth is a combination of Instant ID and ControlNet Depth models.
-
ID Embedding: This part analyzes the input image to capture the person's unique facial features, like eye color, nose shape, etc. It focuses on these defining characteristics (semantic information) rather than the exact location of each feature on the face (spatial information).
-
Lightweight Adapted Module: This module acts like an adapter, allowing the system to use the reference image itself as a visual prompt for the image generation process. The reference image can be any pose image.
-
IdentityNet: This is where the actual image generation happens. It takes the information from the ID embedding (facial characteristics) and combines it with the text prompt to create a new image.
-
ControlNet Depth enables composition transfer by understanding the depth of the input face image. It accurately preserves the person’s face in the new pose (reference image) in the output image.
How to use Insta Depth
-
Input image: Provide a clear image of the person you want to generate variations for. This image is used to capture unique facial features and characteristics of the person.
-
Pose Image: Upload a reference image that represents the pose you want the person in the input image to take. This could be any pose like standing, jumping, sitting, etc.
-
Prompt: Provide a text prompt that describes the final output you envision. For example, if you want the person in the image to appear as if they’re wearing a Wonder Woman costume, your prompt could be “Photo of a woman wearing a Wonder Woman costume”.
Other Popular Models
sdxl-img2img
SDXL Img2Img is used for text-guided image-to-image translation. This model uses the weights from Stable Diffusion to generate new images from an input image using StableDiffusionImg2ImgPipeline from diffusers
sdxl-controlnet
SDXL ControlNet gives unprecedented control over text-to-image generation. SDXL ControlNet models Introduces the concept of conditioning inputs, which provide additional information to guide the image generation process
illusion-diffusion-hq
Monster Labs QrCode ControlNet on top of SD Realistic Vision v5.1
codeformer
CodeFormer is a robust face restoration algorithm for old photos or AI-generated faces.
