What Are the Output Formats of bradautomates/claude-video?

The bradautomates/claude-video tool converts video input into exactly two machine-readable artifacts—JPEG image frames and plain-text transcripts—enabling direct multimodal processing by Claude.

The bradautomates/claude-video repository serves as a specialized interface between video sources and Claude's AI capabilities. Understanding the output formats of claude-video reveals how the tool standardizes visual and auditory data into formats optimized for large language model consumption rather than preserving original video containers.

JPEG Image Frame Extraction

The primary visual output consists of JPEG image frames generated via ffmpeg. According to the source implementation, the tool writes extracted frames using sequential naming patterns—either frame_####.jpg for interval-based sampling or cue_####.jpg when extracting at explicit timestamps.

By default, the scaling algorithm fixes the width at 512 pixels and calculates height proportionally to maintain the original aspect ratio. These are standard JPEG files without proprietary encoding, allowing Claude to render them directly within its multimodal interface.

# Example directory structure after processing

./output/
├── frame_0001.jpg
├── frame_0002.jpg
├── cue_0050.jpg    # Extracted at specific timestamp

└── transcript.txt

Plain-Text Transcript Generation

For audio content, claude-video produces plain-text transcripts through two distinct pathways. When native captions exist, the tool invokes yt-dlp to retrieve WebVTT (VTT) formatted subtitles, then parses these structured files into simple text strings with optional timestamp associations.

If no captions are available, the audio stream routes through Whisper using either the Groq or OpenAI API implementation. The resulting transcription converts speech to unstructured plain text, discarding VTT markup while preserving the semantic content for Claude's text window.

Output Delivery and File Characteristics

Upon completion, the tool prints absolute file paths to the console for both artifact types before handing them to Claude. The output deliberately excludes video containers, binary databases, or proprietary formats—only the standardized JPEG sequence and text transcript remain as deliverables.

Summary

• JPEG frames (frame_####.jpg or cue_####.jpg) at 512px default width extracted via ffmpeg
• Plain-text transcripts parsed from VTT sources or generated via Whisper (Groq/OpenAI APIs)
• Direct console output of absolute file paths with no intermediate binary formats
• Optimized specifically for Claude's multimodal input requirements

Frequently Asked Questions

Does claude-video preserve original video files in its output?

No. The tool deliberately transcodes video into static JPEG frames rather than preserving original video containers, as Claude's multimodal capabilities process individual images more effectively than video streams.

What naming convention does claude-video use for extracted frames?

The tool follows zero-padded four-digit naming patterns, producing files like frame_0001.jpg for sequential frames or cue_####.jpg for timestamp-specific extractions, both as standard JPEG images.

Can claude-video output structured caption formats like SRT or VTT?

While the tool ingests VTT files via yt-dlp when available, it outputs only plain-text transcripts without preserving original markup structures like timestamps, positioning data, or style tags.

Which speech-to-text providers does claude-video support?

The transcription fallback supports Whisper implementations through both Groq and OpenAI APIs when native captions are unavailable in the source video.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}