app.py

import gradio as gr
from transformers import VisionEncoderDecoderModel, AutoTokenizer
from PIL import Image
import io

# Load model
model = LLaVAForVisionTextGeneration.from_pretrained("liuhaotian/LLaVA-1.5-7b")
tokenizer = LLaVATokenizer.from_pretrained("liuhaotian/LLaVA-1.5-7b")

# Function to analyze the image
def analyze_image(image_blob):
    image = Image.open(io.BytesIO(image_blob))
    inputs = tokenizer("Analyze the emotions in this image", return_tensors="pt")
    outputs = model.generate(**inputs, images=image)
    return tokenizer.decode(outputs[0])

# Set up the Gradio interface
iface = gr.Interface(fn=analyze_image, inputs="file", outputs="text")
iface.launch()