Create a Dancing Word‑Cloud Video with Python and AI

In this guide we show how to build a dancing word‑cloud video using Python. The workflow includes downloading a dance clip, splitting it into frames, extracting the human figure with Baidu AI, creating word‑cloud masks, and finally compositing the masks into a new video.

1. Download the source video

Use the you-get tool to fetch any online video (e.g., from Bilibili):

you-get url

2. Split the video into frames

OpenCV reads the video and saves every N‑th frame as an image:

import cv2
vc = cv2.VideoCapture(r'美女跳舞视频.flv')
n = 1
timeF = 10  # save one frame every 10 frames
num = 0
while True:
    ret, frame = vc.read()
    if not ret:
        break
    if n % timeF == 0:
        num += 1
        cv2.imwrite(f'{num}.jpg', frame)
    n += 1
    cv2.waitKey(1)
vc.release()

Sample extracted frame:

3. Person segmentation with Baidu AI

Create a Baidu AI application, obtain AppID , API Key and Secret Key , then call the body‑segmentation API on each frame:

APP_ID = '你的APP_ID'
API_KEY = '你的API_KEY'
SECRET_KEY = '你的SECRET_KEY'
client = AipBodyAnalysis(APP_ID, API_KEY, SECRET_KEY)
path = r'美女跳舞视频'
img_files = os.listdir(r'img')
for num in range(2, len(img_files) + 1):
    img = f'img_{num}.jpg'
    with open(img, 'rb') as fp:
        img_info = fp.read()
    seg_res = client.bodySeg(img_info)
    labelmap = base64.b64decode(seg_res['labelmap'])
    nparr = np.frombuffer(labelmap, np.uint8)
    labelimg = cv2.imdecode(nparr, 1)
    labelimg = cv2.resize(labelimg, (width, height), interpolation=cv2.INTER_NEAREST)
    mask = np.where(labelimg == 1, 255, labelimg)
    cv2.imwrite(f'mask_{num}.png', mask)

Console screenshots of the Baidu AI console (app creation, API keys, and body‑analysis page) are omitted for brevity.

4. Generate word‑cloud masks

Read comments, cut words with jieba, and render a word cloud using each mask image:

for num in range(1, 23):
    with open('comment.txt', 'r') as f:
        text = f.read()
    words = jieba.cut(text)
    word_str = " ".join(words)
    mask = 255 - np.array(Image.open(f'mask_{num}.png'))
    wc = WordCloud(stopwords=STOPWORDS.add('一个'), collocations=False,
                  background_color='white', font_path=r"K:\msyh.ttc",
                  width=400, height=300, random_state=42, mask=mask)
    wc.generate(word_str)
    wc.to_file(f'ciyun_{num}.png')

Example word‑cloud result:

5. Assemble the final video

Combine the generated word‑cloud images into a video with OpenCV:

import cv2
video_address = 'tiaowu.mp4'
fps = 20
img_size = (1080, 1920)
fourcc = cv2.VideoWriter_fourcc('M','P','4','V')
videoWriter = cv2.VideoWriter(video_address, fourcc, fps, img_size)
for num in range(1, 23):
    img_path = f'ciyun_{num}.png'
    frame = cv2.resize(cv2.imread(img_path), img_size)
    videoWriter.write(frame)
videoWriter.release()

Resulting dancing word‑cloud video (GIF preview):

6. Summary

The article demonstrates a complete pipeline: download a dance video, extract frames, segment the dancer using Baidu AI, create word‑cloud masks from comments, and stitch the masks into a new video, providing a visual “dancing word cloud” effect.