Image Editing: Face Migration

Feature Description

Face migration refers to transferring a specific human face to a generated image based on the image generation capability of the image model. When a user provides an image containing a person's face, the model will extract that person's facial features and embed them into the generated image during the image creation process according to the prompt.

Models Supporting Face Migration

• flux-1-schnell

Using the API

The following explains how to use the face migration interface through code.

Step 1: Input API Key and Define Request Function

First, obtain your access token, then define the request function as follows:

python

import requests
import base64
from PIL import Image
from io import BytesIO

API_URL = "https://moark.ai/v1/images/face-migration"
headers = {
    "Authorization": "Bearer <add your token>"
}

def query(payload):
    files = {
        "image": (payload["image"], open(payload["image"], "rb"))
    }
    data = {key: payload[key] for key in payload if key not in files}
    response = requests.post(API_URL, headers=headers, files=files, data=data)
    return response.json()

Step 2: Make a Request and Get Results

The request method is as follows:

python

output = query({
    "model": "flux-1-schnell",
    "image": "liuyifei.png",
    "size": "1024x1024",
    "guidance_scale": 4,
    "num_inference_steps": 4,
    "id_weight": 1,
    "id_timestep_start": 0,
    "prompt": "Portrait painting, delicate sketch style, colored painting, ultra-high-definition, ultra-high pixel count"
})

The prompt in the above code Portrait painting, delicate sketch style, colored painting, ultra-high-definition, ultra-high pixel count requires generating a high-definition, ultra-high pixel count colored sketch portrait. The tutorial uses the following id image:

Parameter Explanations:

• model: Model name, fixed here.
• image: Local image path; as can be seen in the function, it is transmitted in binary form.
• size: Dimensions of the generated image.
• guidance_scale: Prompt guidance coefficient; larger values mean the generated image adheres more closely to the prompt.
• num_inference_steps: Number of steps for image generation; this model can produce excellent results in just 4 steps.
• id_weight: The degree of influence of the ID reference image on the generated image; larger values result in a more accurate facial resemblance, which can be adjusted appropriately.
• id_timestep_start: The step at which face embedding begins during image generation; smaller values mean face generation starts earlier, resulting in a more accurate resemblance, which can be adjusted appropriately.
• prompt: Prompt for image generation; Flux supports long, complex prompts. The more detailed, the better—feel free to be specific.

Step 3: Decode and Save Results

The resulting output is in JSON format as follows:

{
  "data": [
    {
      "b64_json": "<b64_data>"
    }
  ],
  "created": "<id>"
}

The result is a Base64-encoded image, so it needs to be decoded before saving. Unpack and decode as follows:

python

b64data = output['data'][0]['b64_json']
img = Image.open(BytesIO(base64.b64decode(b64data)))
img.save("/path/to/save/res.jpg")
img.show()

The result is shown below, which generated a colored sketch portrait of the specific face as requested:

The complete example code is as follows:

python

import requests
import base64
from PIL import Image
from io import BytesIO

API_URL = "https://ai.gitee.com/v1/images/face-migration"
headers = {
    "Authorization": "Bearer <add your token>"
}

def query(payload):
    files = {
        "image": (payload["image"], open(payload["image"], "rb"))
    }
    data = {key: payload[key] for key in payload if key not in files}
    response = requests.post(API_URL, headers=headers, files=files, data=data)
    return response.json()

output = query({
    "model": "flux-1-schnell",
    "image": "liuyifei.png",
    "size": "1024x1024",
    "guidance_scale": 4,
    "num_inference_steps": 4,
    "id_weight": 1,
    "id_timestep_start": 0,
    "prompt": "a woman holding sign with glowing green text \"Gitee AI\"",
})

data = output["data"][0]["b64_json"]

stream = BytesIO(base64.b64decode(data))
img = Image.open(stream)
img.save("output.png")
img.show()