multi_view.py

import torch
from diffusers import DiffusionPipeline
from PIL import Image
from utils import create_image_grid  # Changed import

class MultiViewDiffusion:
    def __init__(self, device="cuda", model_id = "dylanebert/mvdream"):
        self.device = device
        self.pipeline = DiffusionPipeline.from_pretrained(
            model_id,
            custom_pipeline="dylanebert/multi-view-diffusion",
            torch_dtype=torch.float16,
            trust_remote_code=True,
        ).to(self.device)


    def generate_views(self, prompt, num_views=4, guidance_scale=5, num_inference_steps=30, elevation=0):
        images = self.pipeline(
            prompt,
            guidance_scale=guidance_scale,
            num_inference_steps=num_inference_steps,
            elevation=elevation,
        ).images
        return images

if __name__ == "__main__":
    mv_diff = MultiViewDiffusion()
    images = mv_diff.generate_views("A futuristic city")
    grid = create_image_grid(images)
    grid.save("multi_view_output.png")