import gradio as gr
from transformers import AutoModelForImageTextToText, AutoProcessor
from PIL import Image
import torch

model_id = "Rishabh3365/smolvlm2-blv-500m"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForImageTextToText.from_pretrained(
    model_id, torch_dtype=torch.float32
)
model.eval()

def describe_scene(image, instruction):
    inputs = processor(images=image, text=instruction, return_tensors="pt")
    with torch.no_grad():
        output = model.generate(**inputs, max_new_tokens=200)
    return processor.decode(output[0], skip_special_tokens=True)

gr.Interface(
    fn=describe_scene,
    inputs=[
        gr.Image(sources=["webcam", "upload"], type="pil", label="Capture / Upload"),
        gr.Textbox(value="What do you see? Describe for a blind person.", label="Instruction")
    ],
    outputs=gr.Textbox(label="BLV Navigation Response"),
    title="SmolVLM2 BLV Navigation Assistant",
    description="Captures a scene and describes it for blind and low-vision users."
).launch()