๐ŸŽฌ Frame Arena: Frame by frame comparisons of any videos

๐ŸŽ‰ This tool has been created to celebrate our Wan 2.2 text-to-video and image-to-video endpoints on Replicate. Want to know more? Check out our blog!

  • Upload videos in common formats with the same number of frames (MP4, AVI, MOV, etc.) or use URLs
  • 7 Quality Metrics: SSIM, PSNR, MSE, pHash, Color Histogram, Sharpness + Overall Quality
  • Individual Visualization: Each metric gets its own dedicated plot
  • Real-time Analysis: Navigate frames with live metric updates
  • Smart Comparisons: Understand differences between videos per metric

Perfect for: Analyzing compression effects, processing artifacts, visual quality assessment, and compression algorithm comparisons.

Video 1

Video 2

๐Ÿ“ Example Video Comparisons

Click any example to load video pairs:
Upload Video 1 Upload Video 2

Video 1 - Current Frame

Frame Slider (Left: Video 1, Right: Video 2)

โ—ขโ—ข

Video 2 - Current Frame

0 0

๐Ÿ“Š Metric Analysis

๐Ÿ“Š Metrics Explained

  • SSIM: Structural Similarity (1.0 = identical structure, 0.0 = completely different)
  • PSNR: Peak Signal-to-Noise Ratio in dB (higher = better quality, less noise)
  • MSE: Mean Squared Error (lower = more similar pixel values)
  • pHash: Perceptual Hash similarity (1.0 = visually identical)
  • Color Histogram: Color distribution correlation (1.0 = identical color patterns)
  • Sharpness: Laplacian variance per video (higher = sharper/more detailed images)
  • Overall Quality: Combined metric averaging SSIM, min-max normalized PSNR, and pHash

๐ŸŽฏ Quality Assessment Scale (Research-Based Thresholds)

SSIM Scale (based on human perception studies):

  • ๐ŸŸข Excellent (โ‰ฅ0.9): Visually indistinguishable differences
  • ๐Ÿ”ต Good (โ‰ฅ0.8): Minor visible differences, still high quality
  • ๐ŸŸก Fair (โ‰ฅ0.6): Noticeable differences, acceptable quality
  • ๐Ÿ”ด Poor (<0.6): Significant visible artifacts and differences

PSNR Scale (standard video quality benchmarks):

  • ๐ŸŸข Excellent (โ‰ฅ40dB): Professional broadcast quality
  • ๐Ÿ”ต Good (โ‰ฅ30dB): High consumer video quality
  • ๐ŸŸก Fair (โ‰ฅ20dB): Acceptable for web streaming
  • ๐Ÿ”ด Poor (<20dB): Low quality with visible compression artifacts

MSE Scale (pixel difference thresholds):

  • ๐ŸŸข Very Similar (โ‰ค50): Minimal pixel-level differences
  • ๐Ÿ”ต Similar (โ‰ค100): Small differences, good quality preservation
  • ๐ŸŸก Moderately Different (โ‰ค200): Noticeable but acceptable differences
  • ๐Ÿ”ด Very Different (>200): Significant pixel-level changes

๐Ÿ” Understanding Comparisons

Comparison Analysis: Shows how similar/different the videos are

  • Most metrics indicate similarity - not which video "wins"
  • Higher SSIM/PSNR/pHash/Color = more similar videos
  • Lower MSE = more similar videos

Individual Quality: Shows the quality of each video separately

  • Sharpness comparison shows which video has more detail
  • Significance levels: ๐Ÿ”ด MAJOR (>20%), ๐ŸŸก MODERATE (10-20%), ๐Ÿ”ต MINOR (5-10%), ๐ŸŸข NEGLIGIBLE (<5%)

Overall Quality: Combines multiple metrics to provide a single similarity score

  • Formula: Average of [SSIM + normalized_PSNR + pHash]
  • PSNR Normalization: PSNR values are divided by 50dB and capped at 1.0
  • Range: 0.0 to 1.0 (higher = more similar videos overall)
  • Purpose: Provides a single metric when you need one overall assessment
  • Limitation: Different metrics may disagree; check individual metrics for details