app.py

import os
import requests
import gradio as gr
from src.gradio_demo import SadTalker  
from huggingface_hub import snapshot_download

# 1. Download the Brain
def download_model():
    REPO_ID = 'vinthony/SadTalker-V002rc'
    snapshot_download(repo_id=REPO_ID, local_dir='./checkpoints', local_dir_use_symlinks=True)

download_model()
sad_talker = SadTalker(lazy_load=True)

# 2. The Internal AI Agent Logic
# UPDATE ONLY THIS FUNCTION IN YOUR app.py
def farmer_agent(question):
    # A. Get Advice from Groq (This part is working!)
    groq_url = "https://api.groq.com/openai/v1/chat/completions"
    headers = {"Authorization": "Bearer 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"}
    payload = {
        "model": "llama-3.1-8b-instant",
        "messages": [{"role": "system", "content": "You are a village farmer expert. 2 sentences."},
                     {"role": "user", "content": question}]
    }
    resp = requests.post(groq_url, json=payload, headers=headers).json()
    advice = resp['choices'][0]['message']['content']
    
    # B. Generate the Video (The fix for the AssertionError)
    source_img = "examples/source_image/full_body_1.png" 
    
    # We set use_ref_video=False and ref_info='pose' to bypass the Assertion
    video_path = sad_talker.test(
        source_img, 
        advice, 
        'crop', 
        True, 
        False, 
        1, 
        256, 
        0, 
        'facevid2vid', 
        1.0, 
        False,   # use_ref_video (Changed to False)
        None,    # ref_video
        'pose',  # ref_info (Changed to 'pose' to match logic)
        False, 
        5, 
        True
    )
    return video_path, advice

# 3. Simple 1-Input Interface
with gr.Blocks() as demo:
    gr.Markdown("# 🌾 Farmer GenAI Advisor")
    with gr.Row():
        input_text = gr.Textbox(label="Farmer's Question")
        output_video = gr.Video(label="Advisor Video")
        output_text = gr.Textbox(label="Advice Text")
    
    btn = gr.Button("Ask the Expert")
    btn.click(fn=farmer_agent, inputs=input_text, outputs=[output_video, output_text])

demo.launch(server_name="0.0.0.0", server_port=7860)