Keyframe Extractor

Keyframe Extractor

This skill extracts important frames from videos, identifying visually significant moments, scene changes, and stylistic reference points. Use for style matching, video regeneration, or visual analysis.

When to Use This Skill

• Extracting style reference frames from videos
• Analyzing shot composition and visual style
• Preparing frames for image-to-video generation
• Creating visual mood boards from video
• Documenting video aesthetic for replication
• Extracting frames for prompt generation
• Getting screenshots for storyboards

What This Skill Does

1. Scene Detection: Identifies major scene changes and transitions
2. Keyframe Identification: Selects visually representative frames
3. Frame Extraction: Saves high-quality still images from video
4. Frame Analysis: Provides description of visual elements in each frame
5. Style Documentation: Catalogs visual patterns across frames
6. Metadata Capture: Records timestamp and context for each frame

How to Use

Extract from Video URL

Extract keyframes from this video: [URL]

Get reference frames from: [URL]

Specific Extraction

Extract [N] evenly-spaced frames from: [URL]

Get frames at: 0:05, 0:30, 1:15, 2:00 from [URL]

For Style Reference

Extract keyframes from [URL] for style matching

Instructions

When user requests keyframe extraction:

1. Understand Requirements

   I'll extract keyframes from your video.
   
   **Video Source**: [URL or file path]
   
   **Extraction Method**:
   [ ] Automatic scene detection (AI identifies key moments)
   [ ] Evenly spaced (e.g., every 10 seconds)
   [ ] Specific timestamps (provide times)
   [ ] Smart sampling (balanced across video)
   
   **Number of Frames**: [How many frames to extract]
   
   **Purpose**:
   [ ] Style reference for new video
   [ ] Visual analysis
   [ ] Prompt generation for image recreation
   [ ] Storyboard creation
   [ ] Other: ___________
   

2. Download/Access Video

   ## Video Processing
   
   **Video Details**:
   - Source: [URL or path]
   - Duration: [X:XX]
   - Resolution: [resolution]
   - FPS: [frame rate]
   
   **Extraction Settings**:
   - Method: [Scene detection / Interval / Timestamps]
   - Frame count: [N frames]
   - Output format: PNG (high quality)
   - Resolution: [Match source or specify]
   

3. Extract Frames (Using FFmpeg or similar)

   # Scene detection method
   ffmpeg -i input_video.mp4 -vf "select='gt(scene,0.3)',showinfo" -vsync vfr ./output/assets/keyframes/frame_%04d.png
   
   # Every N seconds method
   ffmpeg -i input_video.mp4 -vf fps=1/10 ./output/assets/keyframes/frame_%04d.png
   
   # Specific timestamps
   ffmpeg -i input_video.mp4 -ss 00:00:05 -vframes 1 ./output/assets/keyframes/frame_001.png
   

4. Analyze Each Frame

   For each extracted frame:

   ## Keyframe Catalog
   
   ### Frame 001
   **Timestamp**: 0:00:05
   **Filename**: `frame_001.png`
   
   **Visual Description**:
   - **Subject**: [What's the main focus]
   - **Composition**: [How elements are arranged]
   - **Shot Type**: [Close-up/Medium/Wide]
   - **Setting**: [Location/background]
   
   **Color Analysis**:
   - **Dominant colors**: [List 3-5 colors]
   - **Color temperature**: [Warm/Cool/Neutral]
   - **Saturation**: [Low/Medium/High]
   - **Mood**: [Emotional quality]
   
   **Lighting**:
   - **Source**: [Natural/Studio/Mixed]
   - **Direction**: [Front/Side/Back]
   - **Quality**: [Soft/Hard]
   - **Mood**: [Bright/Dramatic/Neutral]
   
   **Elements Present**:
   - [ ] Person/People
   - [ ] Product
   - [ ] Text overlay
   - [ ] Graphics/Icons
   - [ ] Logo/Branding
   - [ ] Background elements
   
   **Technical Quality**:
   - Sharpness: [Sharp/Soft/Motion blur]
   - Exposure: [Well-exposed/Overexposed/Underexposed]
   - Depth of field: [Shallow/Deep]
   
   **Use Case**:
   - Style reference for: [What aspect]
   - Prompt potential: [How to recreate]
   - Storyboard position: [Where in narrative]
   
   ---
   
   ### Frame 002
   [Same structure as above]
   
   ---
   
   ### Frame 003
   [Continue for all extracted frames]
   

5. Pattern Analysis

   ## Visual Patterns Across Frames
   
   **Consistent Style Elements**:
   - **Color Palette**: [Colors appearing in multiple frames]
   - **Composition Style**: [Common framing approaches]
   - **Lighting Style**: [Consistent lighting approach]
   - **Visual Tone**: [Overall aesthetic consistency]
   
   **Scene Breakdown**:
   | Frame | Timestamp | Scene Type | Subject | Mood |
   |-------|-----------|------------|---------|------|
   | 001 | 0:05 | Opening | [Subject] | [Mood] |
   | 002 | 0:18 | Intro | [Subject] | [Mood] |
   | 003 | 0:45 | Content | [Subject] | [Mood] |
   | ... | ... | ... | ... | ... |
   
   **Transitions Between Frames**:
   - Frame 1 → 2: [How scene changes]
   - Frame 2 → 3: [How scene changes]
   
   **Visual Progression**:
   - [How visual style evolves through video]
   

6. Style Guide Creation

   ## Visual Style Guide (from extracted frames)
   
   **Overall Aesthetic**: [Description]
   
   **Color Scheme**:
   

   Primary: [Color] (appears in frames: [list]) Secondary: [Color] (appears in frames: [list]) Accent: [Color] (appears in frames: [list])

   
   **Composition Rules**:
   - Rule of thirds: [Yes/No, which frames]
   - Center framing: [Yes/No, which frames]
   - Leading lines: [Yes/No, which frames]
   - Symmetry: [Yes/No, which frames]
   
   **Lighting Characteristics**:
   - Typical setup: [Description]
   - Key light position: [Common pattern]
   - Fill/background: [Treatment]
   - Overall brightness: [Light/Medium/Dark]
   
   **Common Visual Elements**:
   - Logo placement: [Where and how]
   - Text style: [Font, size, position]
   - Graphics: [Style description]
   - Backgrounds: [Typical settings]
   
   **Shot Variety**:
   - Close-ups: [X% of frames]
   - Medium shots: [X% of frames]
   - Wide shots: [X% of frames]
   - B-roll: [X% of frames]
   

7. Generate Image Prompts

   For each frame, create recreation prompt:

   ## Image Generation Prompts
   
   ### Frame 001 Prompt
   

   [Detailed prompt to recreate this frame]

   Style: [aesthetic description] Subject: [main subject] Setting: [background/environment] Lighting: [lighting setup] Colors: [color palette] Mood: [emotional tone] Camera: [shot type and angle] Quality: [technical specs - 4K, photorealistic, etc.]

   
   **Negative prompt**: [Elements to avoid]
   
   ---
   
   ### Frame 002 Prompt
   [Same structure]
   
   ---
   
   [Continue for all frames]
   

8. Organize Output

   ## File Organization
   
   **Directory Structure**:
   

   /output/assets/keyframes/[video-name]/ ├── frame_001.png (timestamp: 0:05) ├── frame_002.png (timestamp: 0:18) ├── frame_003.png (timestamp: 0:45) ├── ... ├── keyframe_analysis.md (this document) └── prompts.txt (all prompts for regeneration)

   
   **Frame Index**:
   - Total frames: [N]
   - Format: PNG
   - Resolution: [WxH]
   - File size: ~[X]MB per frame
   

9. Provide Recommendations

   ## Recommendations for Use
   
   **For Style Matching**:
   - Use frames [X, Y, Z] as primary reference
   - Match color palette from frame [X]
   - Replicate composition style from frame [Y]
   - Use lighting from frame [Z] as template
   
   **For Regeneration**:
   - Frames [X, Y] are best for image-to-video start points
   - Frame [Z] works well for ending shot
   - Transitions needed between: [frame pairs]
   
   **For Storyboarding**:
   - Frame sequence represents: [narrative flow]
   - Use as visual reference for: [storyboard sections]
   - Adapt composition from: [specific frames]
   

10. Create Comparison Grid (Optional)

    ## Visual Grid
    
    [Create a markdown table or description showing frames side-by-side]
    
    | Frame 001 | Frame 002 | Frame 003 |
    |-----------|-----------|-----------|
    | [0:05] | [0:18] | [0:45] |
    | Opening hook | Intro | Main content |
    | Wide shot | Medium | Close-up |
    
    **Visual Evolution**: [How visual style progresses]
    

11. Save All Assets

    • Save frames to: ./output/assets/keyframes/[video-name]/
    • Save analysis to: ./output/assets/keyframes/[video-name]/keyframe_analysis.md
    • Save prompts to: ./output/assets/keyframes/[video-name]/prompts.txt
    • Add thumbnails to: ./output/assets/keyframes/[video-name]/thumbnails/ (optional)

12. Provide Summary

    ✅ Keyframe extraction complete!
    
    **Extracted**: [N] frames from [video name]
    **Saved to**: `./output/assets/keyframes/[video-name]/`
    
    **Frame Summary**:
    - Total frames: [N]
    - Time span: [start] to [end]
    - Resolution: [WxH]
    - Style: [Overall aesthetic]
    
    **Notable Frames**:
    - Frame [X] (timestamp): [Why notable]
    - Frame [Y] (timestamp): [Why notable]
    - Frame [Z] (timestamp): [Why notable]
    
    **Files Created**:
    - [N] PNG frame files
    - keyframe_analysis.md
    - prompts.txt
    
    **Next Steps**:
    1. Review frames and analysis
    2. Use frames for style reference
    3. Generate images: `Generate images from these keyframe prompts`
    4. Create storyboard: `Create storyboard using these frames`
    5. Analyze patterns: `Analyze visual style from these frames`
    
    What would you like to do with these frames?
    

Output Format

Keyframe extraction should produce:

• High-quality frame images: PNG format, original resolution
• Detailed analysis: Visual description for each frame
• Image prompts: Ready to use for regeneration
• Style guide: Patterns and consistency analysis
• Organized files: Clear naming and folder structure

Tips

• Scene detection: Works best for videos with clear scene changes
• Interval method: Good for consistent sampling across video
• Quality over quantity: 5-10 meaningful frames better than 50 random ones
• Timestamp notes: Always record exact timestamp for context
• Prompt detail: More specific prompts = better regeneration results
• Style consistency: Look for patterns across frames, not just individual shots

Technical Notes

# Extract one frame per second
ffmpeg -i input.mp4 -vf fps=1 output_%04d.png

# Extract high-quality frames at specific times
ffmpeg -i input.mp4 -ss 00:01:30 -vframes 1 -q:v 2 output.png

# Extract frames with scene detection (0.3 = sensitivity)
ffmpeg -i input.mp4 -vf "select='gt(scene,0.3)'" -vsync vfr output_%04d.png

# Extract every Nth frame
ffmpeg -i input.mp4 -vf "select='not(mod(n\,30))'" -vsync vfr output_%04d.png

Related Skills

• Use video-analyzer first to understand video structure
• Use prompt-generator to create detailed prompts from frames
• Use batch-image-generator to recreate style from frame prompts
• Use frames as reference when using video-segment-creator