Skip to content
TheChatArchiveThe Chat Archive
o3-mini· OpenAIHugging Face LMArena Human Preference 140K

Create just a bash script to convert an image to AVIF with Image Magick, it should find…

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with gpt-4.1-2025-04-14’s answer.
  1. User

    Create just a bash script to convert an image to AVIF with Image Magick, it should find the right quality value to get a similarity of 99%

  2. Assistant

    Below is an example Bash script that uses ImageMagick’s convert and compare tools to “tune” the AVIF compression quality until the SSIM similarity compared to the input image is at least 99%. (Note that you must have ImageMagick compiled with AVIF (often via libheif) support, and the compare command must support the SSIM metric.) This script uses a simple binary search over quality values from 0 to 100 to find the lowest acceptable quality that gives at least 99% similarity.

    Save the script (for example as to_avif.sh), make it executable (e.g. chmod +x to_avif.sh), and run it with the input and output file names as arguments.


    #!/bin/bash
    set -e
    
    # Check arguments
    if [ "$#" -ne 2 ]; then
      echo "Usage: $0 input_image output_image.avif"
      exit 1
    fi
    
    input="$1"
    output="$2"
    
    # Temporary file for intermediate AVIF conversion
    tmp_avif=$(mktemp --suffix=.avif)
    
    # Desired SSIM similarity (e.g., 0.99 means 99% similarity)
    desired_similarity=0.99
    
    # Set initial binary search range for quality (0-100)
    low=0
    high=100
    best=100
    
    echo "Tuning quality for SSIM >= $desired_similarity..."
    
    while [ "$low" -le "$high" ]; do
      mid=$(( (low + high) / 2 ))
      echo "  Trying quality: $mid"
    
      # Convert the input to AVIF with quality=mid
      convert "$input" -quality "$mid" "$tmp_avif"
    
      # Compare the input and the temporary AVIF using the SSIM metric.
      # Note: compare prints the metric value to stderr.
      similarity=$(compare -metric SSIM "$input" "$tmp_avif" null: 2>&1)
    
      echo "    SSIM similarity: $similarity"
    
      # Check if the similarity meets or exceeds the threshold.
      # The bc tool is used for floating point comparison.
      if [ "$(echo "$similarity >= $desired_similarity" | bc -l)" -eq 1 ]; then
        best=$mid
        # Try to see if a lower quality (smaller file size) still meets the similarity
        high=$(( mid - 1 ))
      else
        low=$(( mid + 1 ))
      fi
    done
    
    echo "Selected quality value: $best"
    echo "Converting $input to AVIF with quality=$best ..."
    
    # Do the final conversion with the found best quality.
    convert "$input" -quality "$best" "$output"
    
    # Clean up the temporary file
    rm "$tmp_avif"
    
    echo "Done! Output image saved as: $output"
    

    How It Works

    1. Input checks and temporary file creation:
      The script expects exactly two arguments (the input image and the desired output file name) and creates a temporary file for the intermediate AVIF conversion.

    2. Binary Search:
      It performs a binary search over the integer quality values from 0 to 100. For each quality value (mid):

      • The script converts the original image to AVIF.
      • It then uses the compare command with the SSIM metric (which prints a number between 0 and 1) comparing the original input and the current AVIF image.
      • If the SSIM is at least 0.99 (i.e. 99% similarity), the script remembers this quality and then continues searching for a possibly lower acceptable quality.
      • If the similarity is too low, it increases the quality value.
    3. Final conversion:
      Once a best quality value is determined, the script runs the conversion a final time to produce the output file.

    4. Cleanup:
      The temporary AVIF file is removed.

    Make sure you have the required versions of ImageMagick and bc installed before using the script. Enjoy your AVIF conversions!

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.