import spaces
import huggingface_hub
huggingface_hub.snapshot_download(
repo_id='h94/IP-Adapter',
allow_patterns=[
'models/**',
'sdxl_models/**',
],
local_dir='./'
)
import gradio as gr
from diffusers import StableDiffusionXLControlNetInpaintPipeline, ControlNetModel
from rembg import remove
from PIL import Image
import torch
from ip_adapter import IPAdapterXL
from ip_adapter.utils import register_cross_attention_hook, get_net_attn_map, attnmaps2images
from PIL import Image, ImageChops, ImageEnhance
import numpy as np
import os
import glob
import torch
import cv2
import argparse
import DPT.util.io
from torchvision.transforms import Compose
from DPT.dpt.models import DPTDepthModel
from DPT.dpt.midas_net import MidasNet_large
from DPT.dpt.transforms import Resize, NormalizeImage, PrepareForNet
"""
Get ZeST Ready
"""
base_model_path = "stabilityai/stable-diffusion-xl-base-1.0"
image_encoder_path = "models/image_encoder"
ip_ckpt = "sdxl_models/ip-adapter_sdxl_vit-h.bin"
controlnet_path = "diffusers/controlnet-depth-sdxl-1.0"
device = "cuda"
torch.cuda.empty_cache()
# load SDXL pipeline
controlnet = ControlNetModel.from_pretrained(controlnet_path, variant="fp16", use_safetensors=True, torch_dtype=torch.float16).to(device)
pipe = StableDiffusionXLControlNetInpaintPipeline.from_pretrained(
base_model_path,
controlnet=controlnet,
use_safetensors=True,
torch_dtype=torch.float16,
add_watermarker=False,
).to(device)
pipe.unet = register_cross_attention_hook(pipe.unet)
ip_model = IPAdapterXL(pipe, image_encoder_path, ip_ckpt, device)
"""
Get Depth Model Ready
"""
model_path = "DPT/weights/dpt_hybrid-midas-501f0c75.pt"
net_w = net_h = 384
model = DPTDepthModel(
path=model_path,
backbone="vitb_rn50_384",
non_negative=True,
enable_attention_hooks=False,
)
normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
transform = Compose(
[
Resize(
net_w,
net_h,
resize_target=None,
keep_aspect_ratio=True,
ensure_multiple_of=32,
resize_method="minimal",
image_interpolation_method=cv2.INTER_CUBIC,
),
normalization,
PrepareForNet(),
]
)
model.eval()
@spaces.GPU()
def infer(input_image, material_exemplar, progress=gr.Progress(track_tqdm=True)):
"""
Perform zero-shot material transfer from a single input image and a material exemplar image.
This function uses a combination of a depth estimation model (DPT), foreground/background separation,
grayscale stylization, and IP-Adapter+ControlNet with Stable Diffusion XL to generate an output image
in which the material style from the exemplar image is applied to the input image's object.
Args:
input_image (PIL.Image): The original image containing the object to which the new material will be applied.
material_exemplar (PIL.Image): A reference image whose material (texture, reflectance, etc.) is to be transferred to the object in the input image.
progress (gradio.Progress, optional): For tracking the progress bar in Gradio UI. Default enables tqdm tracking.
Returns:
PIL.Image: The output image showing the object from `input_image` rendered with the material of `material_exemplar`.
Steps:
1. Compute a depth map from `input_image` using a DPT-based model.
2. Remove the background from the input image to isolate the object and convert it into a grayscale version.
3. Combine and align the input image, depth map, and mask for use with the IP-Adapter + ControlNet SDXL pipeline.
4. Use the `IPAdapterXL.generate()` function to synthesize a new image by guiding generation using:
- material_exemplar for style/material guidance
- input_image's structure/content in grayscale
- the estimated depth map for spatial layout
- the mask for region-specific conditioning (object-only)
5. Return the first image in the generated list as the final material transfer result.
"""
"""
Compute depth map from input_image
"""
img = np.array(input_image)
img_input = transform({"image": img})["image"]
# compute
with torch.no_grad():
sample = torch.from_numpy(img_input).unsqueeze(0)
# if optimize == True and device == torch.device("cuda"):
# sample = sample.to(memory_format=torch.channels_last)
# sample = sample.half()
prediction = model.forward(sample)
prediction = (
torch.nn.functional.interpolate(
prediction.unsqueeze(1),
size=img.shape[:2],
mode="bicubic",
align_corners=False,
)
.squeeze()
.cpu()
.numpy()
)
depth_min = prediction.min()
depth_max = prediction.max()
bits = 2
max_val = (2 ** (8 * bits)) - 1
if depth_max - depth_min > np.finfo("float").eps:
out = max_val * (prediction - depth_min) / (depth_max - depth_min)
else:
out = np.zeros(prediction.shape, dtype=depth.dtype)
out = (out / 256).astype('uint8')
depth_map = Image.fromarray(out).resize((1024, 1024))
"""
Process foreground decolored image
"""
rm_bg = remove(input_image)
target_mask = rm_bg.convert("RGB").point(lambda x: 0 if x < 1 else 255).convert('L').convert('RGB')
mask_target_img = ImageChops.lighter(input_image, target_mask)
invert_target_mask = ImageChops.invert(target_mask)
gray_target_image = input_image.convert('L').convert('RGB')
gray_target_image = ImageEnhance.Brightness(gray_target_image)
factor = 1.0 # Try adjusting this to get the desired brightness
gray_target_image = gray_target_image.enhance(factor)
grayscale_img = ImageChops.darker(gray_target_image, target_mask)
img_black_mask = ImageChops.darker(input_image, invert_target_mask)
grayscale_init_img = ImageChops.lighter(img_black_mask, grayscale_img)
init_img = grayscale_init_img
"""
Process material exemplar and resize all images
"""
ip_image = material_exemplar.resize((1024, 1024))
init_img = init_img.resize((1024,1024))
mask = target_mask.resize((1024, 1024))
num_samples = 1
images = ip_model.generate(pil_image=ip_image, image=init_img, control_image=depth_map, mask_image=mask, controlnet_conditioning_scale=0.9, num_samples=num_samples, num_inference_steps=30, seed=42)
return images[0]
css = """
#col-container{
margin: 0 auto;
max-width: 960px;
}
"""
with gr.Blocks(css=css) as demo:
with gr.Column(elem_id="col-container"):
gr.Markdown("""
# ZeST: Zero-Shot Material Transfer from a Single Image
Upload two images -- input image and material exemplar. (both 1024*1024 for better results)
ZeST extracts the material from the exemplar and cast it onto the input image following the original lighting cues.
""")
gr.HTML("""
""")
with gr.Row():
with gr.Column():
with gr.Row():
input_image = gr.Image(type="pil", label="input image")
input_image2 = gr.Image(type="pil", label = "material examplar")
submit_btn = gr.Button("Submit")
gr.Examples(
examples = [["demo_assets/input_imgs/pumpkin.png", "demo_assets/material_exemplars/cup_glaze.png"]],
inputs = [input_image, input_image2]
)
with gr.Column():
output_image = gr.Image(label="transfer result")
submit_btn.click(fn=infer, inputs=[input_image, input_image2], outputs=[output_image])
demo.queue().launch(show_error=True, ssr_mode=False, mcp_server=True)
