Generative-AI Hackathon -- Character.XYZ :

• The AI model generates avatar faces using a custom stable diffusion model
• Deploy the model to Hugging Face
• Test the output using Postman

Expected :

• Code implementation :

  • def generate_avatar_face(image_path):
    
    • Parameters:
      - image_path (str): The path to the input image.
      
    • Returns:
      - avatar_face (PIL.Image): The generated avatar face.
      

• Test :
  

  • Using Postman:

    •  Base URL     : https://example.com/api/
      

    •  Request      : POST /avatar-face
      

    •  Content-Type : png/jpeg
      

Actual :

• Code Implementation :

  • def generate_avatar_face(prompt, negative_prompt, num_samples, num_inference_steps, guidance_scale, strength, image_url):
    
    • Parameters:
      
      • image_path (str) : The path to the input image . [Mandatory]
        
      • prompt (str) : Description of image . [Optional]
        
      • negative_prompt(str) : The description which shouldn't be in image . [Optional]
        
      • num_samples (num) : Number of output images to be produced . [Optional]
        
      • num_inference_steps(num): Number of steps to process the image . [Optional]
        
      • guidance_scale(num) : Describes the freedom of model to follow the prompt . [Optional]
        
      • strength (num) : Describes the noise percentage to be added to original image . [Optional]
        
    • Returns:
      
      • avatar_face (.png[Image]): The generated avatar face.
        

• Deployment & Testing :

  • Using Gradio :

    • Gradio takes care of
        -  1.Deployment on Hugging-Face
        -  2.API Testing
        -  3.Front-End Web UI
        -  4.So, No need of postman-testing
      

• Extra-Content :

   -    Stable-Diffusion-webUI using [AUTOMATIC1111/stable-diffusion-webui](https://github.com/AUTOMATIC1111/stable-diffusion-webui.git)
   -    Very powerful and most efficient stable-diffusion platform
   -    Control-Nets, Multi-Controlnets, Openpose editor and many many extra features can be operated on sd with this
   -    Requries a large of free space (70GB), 8GB VRAM (std.) on NVIDIA GPUs.