Img2Img2 seems make some bgcolor to blur whole image

I have a problem when using diffuser to Img2Img,what i have done is following:

  1. First use ffmpeg to convert the reference video into an image;
  2. Then use GFPGAN to repair the picture in high definition;
  3. Use Img2Img to transform the character’s image;
  4. Use ffmpeg to merge images into videos

the result seems blink,you can see the result video

what’s did i miss?
my code is

from diffusers import StableDiffusionImg2ImgPipeline
from diffusers import DPMSolverMultistepScheduler
import torch
from PIL import Image
import os


model_path = 'realisticVisionV60B1_v51VAE.safetensors'
pipe = StableDiffusionImg2ImgPipeline.from_single_file(
    model_path,
    torch_dtype=torch.float32,
    safety_check=True,
    )
pipe = pipe.to("cuda")


#prompt = "a woman in a yellow suit is on a news set with a laptop computer in front of her and a blue background"
prompt = "a woman in a yellow suit is on a news set with a laptop computer in front of her"
negtive_prompt = "earrings,wrinkle,forehead lines,blurry,low quality,low resolution"

seed_id = 3563236128
num_inference_steps = 10
guidance_scale = 5
strength = 0.3
initial_noise_multiplier = 1


generator = torch.Generator("cuda").manual_seed(seed_id)

pipe.scheduler = DPMSolverMultistepScheduler.from_config(
        pipe.scheduler.config, 
        algorithm_type="sde-dpmsolver++",
        use_karras_sigmas=True,
        euler_at_finale=True,
        )

Input_Folder = "SourcePngs"

#
for root, dirs, files in os.walk(Input_Folder):
    for file in files:
        if file.endswith(".png"):
            ImgName = file.split(".")[0]
            ImgFullPath = os.path.join(root, file)
            #
            Img = Image.open(ImgFullPath).convert("RGB")
            #
            image = pipe(
                    prompt=prompt,
                    negtive_prompt=negtive_prompt,
                    generator=generator,
                    num_inference_steps=num_inference_steps,
                    width=1100,
                    height=1200,
                    num_images_per_prompt=1,
                    guidance_scale=guidance_scale,
                    strength=strength,
                    initial_noise_multiplier=initial_noise_multiplier,
                    
                    image=Img).images[0]  
            image.save(ImgFullPath)
            print("Image Saved to :", ImgFullPath)