tiktok analyzer

Developers building social listening tools or research agents often hit a wall: LLMs cannot natively browse TikTok or watch video files. This skill bridges that gap by programmatically deconstructing short-form video into machine-readable data. It extracts high-resolution frames, generates a timestamped audio transcript using Whisper, and pulls full metadata including captions and engagement metrics.

What it does

• Video Deconstruction samples frames at regular intervals and extracts audio for processing.
• Speech-to-Text generates timestamped transcripts using OpenAI Whisper models.
• Metadata Extraction retrieves captions, hashtags, sound info, and engagement counts.
• OCR and Visual Analysis identifies burned-in text, products, and visual demos within frames.
• Cross-Verification aligns transcript segments with visual frames for high-accuracy fact extraction.

How it works

1. Extract Data by running the provided Python script with a TikTok URL or local file path to generate an output directory.
2. Ingest Metadata by reading the generated metadata.json and transcript.txt files for initial context.
3. Analyze Frames using the Read tool to inspect sampled images, mapping filenames to timestamps for visual evidence.
4. Synthesize Findings by cross-referencing what was said, shown, and claimed in the caption to answer specific queries.

Frameworks & tools

The skill utilizes Python, yt-dlp for video retrieval, OpenAI Whisper for transcription, and ffmpeg (via imageio-ffmpeg) for frame sampling.

Why this beats prompting it yourself

Standard prompting fails because LLMs cannot access TikTok URLs or "see" video content directly. This skill handles the heavy lifting of environment setup, video downloading, and multi-modal data alignment so you can query video content as easily as text.

Use cases

• Content Auditing to extract facts, claims, and product mentions for brand safety.
• Trend Analysis by summarizing key points and themes across multiple viral videos.
• Fact-Checking by isolating specific creator claims for external verification.
• Accessibility by generating detailed text descriptions and transcripts for short-form media.

Known limitations

Does not support photo carousels or private/region-blocked videos. Whisper may hallucinate brand names or numbers, requiring visual cross-referencing. Initial run requires model download time.