Create a Dancing Word Cloud from Bilibili Videos with Python – Full Step‑by‑Step Guide

Import Required Modules

Install necessary Python packages (lxml, requests, pandas, numpy, you-get, opencv-python, jieba, fake_useragent, matplotlib, moviepy, etc.) using a script that runs pip install for each library.

import os, time
libs = {"lxml","requests","pandas","numpy","you-get","opencv-python","fake_useragent","matplotlib","moviepy"}
for lib in libs:
    os.system(f"pip3 install -i https://pypi.doubanio.com/simple/ {lib}")
    print(lib + "下载成功")

Download Bilibili Video

Use you-get to download the target video from Bilibili.

pip install you-get
you-get -i https://www.bilibili.com/video/BV11C4y1h7nX

Split Video into Frames

Read the video with OpenCV and save each frame as an image.

import cv2
cap = cv2.VideoCapture(r"video.flv")
num = 1
while True:
    ret, frame = cap.read()
    if ret:
        cv2.imwrite(f"./pictures/img_{num}.jpg", frame)
        num += 1
    else:
        break
cap.release()

Human Segmentation with Baidu AI

Call Baidu Body Analysis API to obtain a mask for each frame and save the segmented foreground.

from aip import AipBodyAnalysis
import base64, numpy as np, cv2, os
client = AipBodyAnalysis(APP_ID, API_KEY, SECRET_KEY)
for img_path in os.listdir('./pictures'):
    with open(os.path.join('./pictures', img_path), 'rb') as fp:
        img_info = fp.read()
    seg_res = client.bodySeg(img_info)
    labelmap = base64.b64decode(seg_res['labelmap'])
    labelimg = cv2.imdecode(np.frombuffer(labelmap, np.uint8), 1)
    mask = np.where(labelimg == 1, 255, 0)
    cv2.imwrite(f"./mask_img/mask_{img_path}.png", mask)

Scrape Bilibili Danmu (Comments)

Fetch historical danmu data for a given date range, clean the text, and store it in an Excel file.

def Grab_barrage(date):
    headers = {
        "origin": "https://www.bilibili.com",
        "referer": "https://www.bilibili.com/video/BV1jZ4y1K78N",
        "cookie": "",
        "user-agent": ua.random()
    }
    params = {"type": 1, "oid": "222413092", "date": date}
    r = requests.get(url, params=params, headers=headers)
    comments = re.findall(r'<d p=".*?">(.*?)</d>', r.text)
    # Append comments to DataFrame and save to Excel

Generate Word Cloud

Load the danmu text, perform jieba segmentation, remove stop words, and create a word‑cloud image with a custom mask.

from wordcloud import WordCloud
import jieba, collections, numpy as np, cv2
text = open('barrages.txt', encoding='utf-8').read()
words = jieba.cut(text, cut_all=True)
stop_words = set(open('stoplist.txt', encoding='utf-8').read().split('
'))
filtered = [w for w in words if w not in stop_words and len(w) > 1]
freq = collections.Counter(filtered)
mask = np.array(cv2.imread('mask.png'))
wc = WordCloud(mask=mask, background_color='black', font_path='simhei.ttf').generate_from_frequencies(freq)
wc.to_file('wordcloud.png')

Compose Final Video

Combine the generated word‑cloud images into a video with OpenCV, then add background music using moviepy.

import cv2
videoWriter = cv2.VideoWriter('result.mp4', cv2.VideoWriter_fourcc('M','P','4','V'), 30, (1920,1080))
for i in range(1, 89):
    frame = cv2.imread(f'./wordcloud/wordcloud_{i}.png')
    videoWriter.write(cv2.resize(frame, (1920,1080)))
videoWriter.release()

import moviepy.editor as mpy
clip = mpy.VideoFileClip('result.mp4')
audio = mpy.AudioFileClip('song.mp3').subclip(0,25)
final = clip.set_audio(audio)
final.write_videofile('final_video.mp4')

Result

The completed video shows a dancing word cloud synchronized with background music, demonstrating the full pipeline from video download to AI‑based segmentation and visual storytelling.