|
import torch |
|
import kiui |
|
import numpy as np |
|
import argparse |
|
from pipeline import MVDreamPipeline |
|
|
|
pipe = MVDreamPipeline.from_pretrained( |
|
|
|
"ashawkey/imagedream-ipmv-diffusers", |
|
torch_dtype=torch.float16, |
|
trust_remote_code=True, |
|
) |
|
pipe = pipe.to("cuda") |
|
|
|
|
|
parser = argparse.ArgumentParser(description="ImageDream") |
|
parser.add_argument("image", type=str, default='data/anya_rgba.png') |
|
parser.add_argument("--prompt", type=str, default="") |
|
args = parser.parse_args() |
|
|
|
for i in range(5): |
|
input_image = kiui.read_image(args.image, mode='float') |
|
image = pipe(args.prompt, input_image, guidance_scale=5, num_inference_steps=30, elevation=0) |
|
grid = np.concatenate( |
|
[ |
|
np.concatenate([image[0], image[2]], axis=0), |
|
np.concatenate([image[1], image[3]], axis=0), |
|
], |
|
axis=1, |
|
) |
|
|
|
kiui.write_image(f'test_imagedream_{i}.jpg', grid) |
|
|