Video Encoding Comparison Tool

This project provides a Bash + HTML-based visual framework to compare the quality of multiple video encodings. It lets you extract reference frames from .mp4 files, display them in carousels, and rate them manually by sharpness, pixelation, and color fidelity. Summary scores are then computed automatically.

---

📦 Features

• Extract 10 consistent random or fixed frames from each video
• Generate a standalone HTML comparison interface
• Rate each frame interactively (ratings are saved to localStorage)
• View average scores per video and per criterion
• See codec, resolution, frame rate, pixel format, bitrate and file size for each encoding
• Play with the Live demo

---

🛠️ Requirements

• Linux or WSL2
• Bash
• ffmpeg
• ffprobe

Ensure ffmpeg and ffprobe are accessible from your shell. Otherwise, install via your package manager:

sudo apt install ffmpeg

---

🚀 Usage

1. Place all .mp4 files you want to compare into a folder (e.g. bigbuckbunny/)
2. Run the script:

chmod +x extract_reference_frames.sh
./extract_reference_frames.sh

3. Follow the prompts:

   • Choose your source folder and output directory
   • Choose whether to extract frames and/or generate HTML
   • Choose extraction mode (fixed or consistent random timestamps)
   • Optionally overwrite existing frames without confirmation

4. Open the generated HTML file (e.g. bigbuckbunny/frames/compare.html) in your browser.

---

📊 How to interpret results

• Click stars to rate each frame's Sharpness, Pixelation, and Color
• Ratings are saved per image locally in your browser (localStorage)
• Average scores are automatically calculated per video
• Each video displays:
  • File size (e.g. 1.5G)
  • Codec details (e.g. H264 Main @ 1920x1080 / 30 fps / yuv420p / 19306 kbps)

---

📁 Output structure

bigbuckbunny/
├── video1.mp4
├── video2.mp4
└── frames/
    ├── video1/
    │   ├── frame_01.png
    │   └── ...
    ├── video2/
    │   └── ...
    └── compare.html

---

📖 Notes

• Ratings are preserved as long as localStorage is not cleared in the browser.
• You can regenerate the HTML file safely without losing your ratings.
• All file paths are relative and work offline.

---

🐰 Demo with Big Buck Bunny video and comparison table of Adobe Media Encoder 2025 presets vs ffmpeg direct export

Big Buck Bunny is an animation movie commonly used to test video playback since it is royalty-free. For this comparison, I have downloaded the 1080p 30fps version which is 264MB.

Then, using Adobe Media Encoder 2025, I have exported a version with all the available presets.

I ended up with the stats below, and 18 GB of videos to watch for find out which preset works the best for my needs.

Preset	Size	Encoding duration
Original file	264M	n/a
Match Source - Adaptive High Bitrate	186M	00:03
Match Source - Adaptive Medium Bitrate	177M	00:03
Match Source - Adaptive Low Bitrate	172M	00:04
H264 - Match Source - 2020	777M	00:10
H264 - Match Source - HLG	777M	00:09
H264 - Match Source - PQ	771M	00:10
Adobe Stock HD with Audio (20Mbps)	1.5G	00:10
Adobe Stock UHD with Audio (40Mbps)	3.0G	00:35
Facebook 720p HD	881M	00:06
Facebook 1080p Full HD	962M	00:03
High Quality 720p HD	771M	00:06
High Quality 1080p HD	1.5G	00:03
Mobile Device 1080p HD	651M	00:03
Twitter 720p HD	180M	00:06
Twitter 1080p Full HD	767M	00:03
Vimeo 720p HD	641M	00:06
Vimeo 1080p Full HD	1.3G	00:03
YouTube 720p HD	1.1G	00:06
YouTube 1080p Full HD	1.2G	00:03

I created a script to extract a few frames in all versions of the video, present them in a synchronized carousel, with a simple notation system.

Go to the demo page now: https://germain-italic.github.io/video-encoding-comparison/bigbuckbunny/frames/compare.html

Use the stars to rate each frame of each preset:

Read the Rating Summary at the bottom of the page.

NB: the rating will work even if you don't rate all images.

Personal ratings and analysis:

Preset	Filesize	Duration	Sharpness	Pixelation	Color	Average	Quality/Size Ratio (%)
Original file	264M	n/a	5	5	5	5	100%
H264 - Match Source - 2020	777M	00:10	4.71	4.43	5.00	4.71	32%
Vimeo 1080p Full HD	1.3G	00:03	4.57	4.43	5.00	4.67	19%
Adobe Stock UHD with Audio (40Mbps)	3.0G	00:35	4.43	4.71	4.86	4.67	10%
Facebook 1080p Full HD	962M	00:03	4.57	4.29	5.00	4.60	26%
YouTube 1080p Full HD	1.2G	00:03	4.29	4.29	5.00	4.52	19%
Adobe Stock HD with Audio (20Mbps)	1.5G	00:10	4.29	4.14	5.00	4.48	16%
High Quality 1080p HD	1.5G	00:03	4.43	4.14	4.86	4.48	16%
Mobile Device 1080p HD	651M	00:03	4.43	4.00	5.00	4.48	37%
High Quality 720p HD	771M	00:06	3.86	3.86	4.86	4.19	29%
YouTube 720p HD	1.1G	00:06	4.14	3.86	4.38	4.14	20%
Vimeo 720p HD	641M	00:06	4.00	3.43	4.86	4.10	34%
Match Source - Adaptive Medium Bitrate	177M	00:03	4.14	3.14	4.86	4.05	77%
Facebook 720p HD	881M	00:06	3.71	3.57	4.83	4.00	24%
Match Source - Adaptive High Bitrate	186M	00:03	3.57	3.14	4.71	3.81	69%
Match Source - Adaptive Low Bitrate	172M	00:04	3.57	3.00	4.71	3.76	74%
Twitter 1080p Full HD	767M	00:03	3.14	2.43	4.29	3.29	23%
H264 - Match Source - HLG	777M	00:09	3.29	4.14	1.50	3.05	21%
H264 - Match Source - PQ	771M	00:10	2.86	3.57	1.00	2.48	17%
Twitter 720p HD	180M	00:06	1.43	1.14	4.00	2.19	65%

Mathematical formula

Quality/Size Ratio = (Quality_preset / Quality_original) × (Size_original / Size_preset) × 100

Where:

• Quality_preset = Average quality score of the preset
• Quality_original = 5 (reference quality)
• Size_original = 264M (reference file size)
• Size_preset = File size of the preset

This formula ensures that:

• Higher quality increases the ratio
• Smaller file size increases the ratio
• The original file has a ratio of exactly 100%
• No preset can exceed 100% since quality cannot exceed the original (5/5)

Analysis results

Best quality/size ratios:

• Match Source - Adaptive Medium Bitrate: 77% (quality 4.05, 177M)
• Match Source - Adaptive Low Bitrate: 74% (quality 3.76, 172M)
• Match Source - Adaptive High Bitrate: 69% (quality 3.81, 186M)

Notable ratios:

• Twitter 720p HD: 65% (quality 2.19, 180M) - very lightweight compensates for low quality
• Mobile Device 1080p HD: 37% (quality 4.48, 651M) - good quality but heavier
• Vimeo 720p HD: 34% (quality 4.10, 641M) - decent balance

Least efficient:

• Adobe Stock UHD with Audio (40Mbps): 10% - highest quality but extremely heavy
• High Quality 1080p HD and Adobe Stock HD: 16% - good quality but very heavy

The "Match Source - Adaptive" presets offer the best efficiency by maintaining reasonable quality while significantly reducing file size.

Thoughts on the AME presets and steps forward with ffmpeg direct export

I realized that all presets exporting files with sufficient quality ("Average" column > 4.5) generate files 3 to 5 times heavier than the original!

As an alternative to AME, here's a test to export with ffmpeg directly:

ffmpeg -i bbb_sunflower_1080p_30fps_normal.mp4 \
  -c:v libx264 -preset slower -crf 23 \
  -tune animation \
  -c:a aac -b:a 160k \
  -map_metadata -1 -movflags +faststart \
  bbb_sunflower_1080p_30fps_normal_crf23_optimized.mp4

This is producing a 230MB video. The encoding process takes 23 minutes.
The quality is very close to the original video according to my visual ratings of still frames in the comparison tool, and the filesize is actually a bit lower.

Another test with crf 21 ends up with 283MB barely any noticeable difference on the test video.

Here is the breakdown:

Parameter	Description	Reason for Use
-i filename.mp4	Specifies the input video file	Source video to be re-encoded
-c:v libx264	Uses the x264 encoder for video	High-quality, widely supported H.264 encoding
-preset slower	Sets the encoding speed/efficiency tradeoff	Slower preset improves compression efficiency, reducing file size
-crf 21	Sets Constant Rate Factor for video quality (lower = better)	CRF 21 is close to original while significantly reducing file size
-tune film (optional)	Adapts settings to specific content types	e.g. film, animation, grain, stillimage, fastdecode, zerolatency
-c:a aac	Sets audio codec to AAC	Good compression and compatibility for audio
-b:a 160k	Sets audio bitrate	Lower than original to reduce size while maintaining acceptable quality
-map_metadata -1	Removes all metadata from output	Avoids unnecessary extra bytes, ensures clean output
-movflags +faststart	Optimizes file for web streaming by moving metadata to the beginning	Enables playback before full download
filename_crf21_optimized.mp4	Output file	Final optimized video

Tunable variables

Variable	Purpose	Typical Values	Effect
-crf	Controls video quality and size	18–28	Lower = better quality, larger size. 23 is default, 18 is near-lossless
-preset	Compression efficiency vs speed	ultrafast → placebo	Slower = better compression, smaller file. Affects encoding time
-b:a	Sets audio quality and size	96k–320k	Lower = smaller file, reduced audio fidelity
-tune	Adjusts encoding to content type	film, animation, grain, stillimage, fastdecode, zerolatency	Fine-tunes encoder settings for specific use cases (optional)

Presets

As mentioned in the documentation the values for preset are:

• ultrafast
• superfast
• veryfast
• faster
• fast
• medium – default preset
• slow
• slower
• veryslow
• placebo

Also worth mentioning:

You can optionally use -tune to change settings based upon the specifics of your input. Current tunings include:

• film – use for high quality movie content; lowers deblocking
• animation – good for cartoons; uses higher deblocking and more reference frames
• grain – preserves the grain structure in old, grainy film material
• stillimage – good for slideshow-like content
• fastdecode – allows faster decoding by disabling certain filters
• zerolatency – good for fast encoding and low-latency streaming

To list all possible internal presets and tunes:

ffmpeg -hide_banner -f lavfi -i nullsrc -c:v libx264 -preset help -f mp4 -

---

🪪 License

MIT License

© 2025 Germain-Italic. Free for personal and commercial use with attribution.

Big Buck Bunny © 2008, Blender Foundation under the Creative Commons Attribution 3.0 license.