ClawCut 🦞✂️#

AI short video generator: topic → script → video with native voice.

Pipeline#

1. Script generation — Gemini 3 Pro generates 9-scene screenplay (Chinese narration + English visual descriptions)
2. Nine-grid image — Gemini 3 Pro Image creates character consistency reference (supports up to 14 input images)
3. Video generation — Veo 3.1 generates 9 video clips concurrently with native Chinese speech
4. Post-processing — Silence trimming + ffmpeg concat into final video

Prerequisites#

• Google Cloud project with Vertex AI enabled
• Service account JSON with Vertex AI User role
• ffmpeg binary installed
• Python 3.11+

Setup#

bash

# Create project from skill scripts
mkdir -p clawcut && cp scripts/*.py scripts/requirements.txt clawcut/
cp assets/.env.example clawcut/.env
cd clawcut

# Create venv and install deps
python3.11 -m venv venv
source venv/bin/activate
pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple

# Configure environment
# Edit .env with your values:
#   GOOGLE_APPLICATION_CREDENTIALS=path/to/service-account.json
#   VERTEX_PROJECT=your-gcp-project-id
#   VERTEX_LOCATION=us-central1
#   FFMPEG_PATH=/usr/local/bin/ffmpeg

Usage#

Gradio UI#

bash

source venv/bin/activate
python3 app.py
# Opens at http://localhost:7860

Modes#

• Topic mode — Enter a topic, generate full video from scratch
• Video imitation — Upload reference video, analyze style and generate matching content
• Multi-image reference — Upload up to 14 images for character consistency

Models (all Vertex AI paid)#

• Script: gemini-3-pro-preview
• Image: gemini-3-pro-image-preview
• Video: veo-3.1-generate-001

Key Features#

• 9-way concurrent video generation (~3 min total)
• Checkpoint/resume (skips existing files)
• Silence trimming (ffmpeg silencedetect)
• Video style imitation from reference
• Up to 14 reference images for character consistency
• All credentials via environment variables (zero hardcoded secrets)