[Auto]

21/12

2026-07-14·13:09·No.943

LOSSY JPG/Webp/JXL/AVIF

CVVDP scores tested: ~9.0 and ~9.5
CVVDP was mainly trained on photographic images so only those will be tested.
My subjective CVVDP interpretation is that ~9.0 = medium fidelity and ~9.5 = high fidelity
The display I'm using is a 27 inch Acer 1080p monitor (~82 PPI), viewing distance is ~20 inches.

CVVDP parameters and info:

cvvdp -t out.png -r in.png --device cpu -d standard_fhd -m cvvdp
[INFO] Running on device: cpu
[INFO] When reporting metric results, please include the following information:
[INFO] "ColorVideoVDP v0.5.6, 37.84 [pix/deg], Lpeak=200, Lblack=0.2, Lrefl=0.3979 [cd/m^2], (standard_fhd)"
ColorVideoVDP v0.5.6 (released on June 4, 2026)

I plan on testing 12 photogramic images, sourced from high quality wikipedia JPGs resized to 50% res PNGs.
I thought about using FCVVDP since that's faster but the slower python one is the most accurate.

JPG encoding parameters and version:

cjpeg -baseline -sample 1x1,1x1,1x1 -quality X in.ppm > out.jpg
Independent JPEG Group's CJPEG, version 6b 27-Mar-1998

Webp encoding parameters and version:

cwebp -m 6 -af -q X in.png -o out.webp
libwebp v1.6.0 (released on June 30, 2025)

JXL encoding parameters and version:

cjxl in.png out.jxl -e 10 -q X
libjxl v0.12.0 (released on July 1, 2026)

AVIF encoding parameters and version:

avifenc -y 420 -d 10 -s 0 -q X in.png out.avif
libavif v1.4.2 (released on May 26, 2026)

Image source for picrel being tested: https://en.wikipedia.org/wiki/File:Crested_myna,_Osaka,_Japan.jpg

Next I'll post my bash script into 3 sections for easier debugging/improvements later for myself and others.

2026-07-14·13:14·No.944·

>>951

Part 1/3 of bash script. Base image encoding params, asks for image codec -q and encodes/decodes to temp PNG files for CVVDP.

#!/bin/bash

jpgbase="-baseline -sample 1x1,1x1,1x1 -quality" && read -p "jpg quality: " jpgq
webpbas="-m 6 -af -q" &&			    read -p "cwebp quality: " webpq
jxlbase="-e 10 -q" &&				    read -p "jxl quality: " jxlq
avifbas="-y 420 -d 10 -s 0 -q" &&		    read -p "avifenc quality: " avifq && echo ""
CVVDPWIN="/mnt/c/Users/1337/AppData/Local/Programs/Python/Python313/Scripts/cvvdp.exe"


# JPG enc/dec
ffmpeg -hide_banner -loglevel quiet -i in.png -y png.ppm && cjpeg $jpgbase $jpgq png.ppm > out.jpg
ffmpeg -hide_banner -loglevel quiet -i out.jpg -y jpg.png

# Webp, JXL, AVIF enc/dec
cwebp $webpbas $webpq in.png -o out.webp 2>&1 | head -n 5 && dwebp -quiet out.webp -o webp.png
cjxl in.png out.jxl $jxlbase $jxlq --quiet && djxl out.jxl jxl.png --quiet
avifenc $avifbas $avifq in.png out.avif &>/dev/null && avifdec out.avif avif.png &>/dev/null

2026-07-14·13:15·No.945

Part 2/3 of bash script. CVVDP/filesize calculated and encoding parameters echo'd.

# Display CVVDP, filesize, and parameters for JPG
jpgcvvdp=$($CVVDPWIN -t jpg.png -r in.png --device cpu -d standard_fhd -m cvvdp 2>&1 \
| awk -F'=' '/cvvdp=/{gsub(/ \[JOD\]/,"",$2); printf "%.2f",$2}')
jpgsize=$(awk 'BEGIN{printf "%.1f",'$(stat -c%s out.jpg)'/1024}') && echo ""
echo "JPG CVVDP= $jpgcvvdp Fsize= $jpgsize KB [ cjpeg $jpgbase $jpgq ]" && echo ""

# Display CVVDP, filesize, and parameters for Webp
webpcvvdp=$($CVVDPWIN -t webp.png -r in.png --device cpu -d standard_fhd -m cvvdp 2>&1 \
| awk -F'=' '/cvvdp=/{gsub(/ \[JOD\]/,"",$2); printf "%.2f",$2}')
webpsize=$(awk 'BEGIN{printf "%.1f",'$(stat -c%s out.webp)'/1024}')
echo "Webp CVVDP= $webpcvvdp Fsize= $webpsize KB [ cwebp $webpbas $webpq ]" && echo ""

# Display CVVDP, filesize, and parameters for JXL
jxlcvvdp=$($CVVDPWIN -t jxl.png -r in.png --device cpu -d standard_fhd -m cvvdp 2>&1 \
| awk -F'=' '/cvvdp=/{gsub(/ \[JOD\]/,"",$2); printf "%.2f",$2}')
jxlsize=$(awk 'BEGIN{printf "%.1f",'$(stat -c%s out.jxl)'/1024}')
echo "JXL CVVDP= $jxlcvvdp Fsize= $jxlsize KB [ cjxl $jxlbase $jxlq ]" && echo ""

# Display CVVDP, filesize, and parameters for AVIF
avifcvvdp=$($CVVDPWIN -t avif.png -r in.png --device cpu -d standard_fhd -m cvvdp 2>&1 \
| awk -F'=' '/cvvdp=/{gsub(/ \[JOD\]/,"",$2); printf "%.2f",$2}')
avifsize=$(awk 'BEGIN{printf "%.1f",'$(stat -c%s out.avif)'/1024}')
echo "AVIF CVVDP= $avifcvvdp Fsize= $avifsize KB [ avifenc $avifbas $avifq]" && echo ""

2026-07-14·13:17·No.946

Part 3/3 of bash script. Diminishing returns calculated with AVIF=winner and JPG=loser. Webp ranked against AVIF and JXL.
Output files also renamed to reflect CVVDP score, filesize, and parameters used. Important for the flicker tests later.

# Diminishing returns comparison
webp_to_avif=$(awk "BEGIN {printf \"%.1f\", (($jpgsize-$webpsize)/($jpgsize-$avifsize))*100}")
webp_to_jxl=$(awk "BEGIN {printf \"%.1f\", (($jpgsize-$webpsize)/($jpgsize-$jxlsize))*100}")
echo "" && echo -e "\e[32mWebP compression efficiency relative to [ JXL:  $webp_to_jxl % ] & [ AVIF: $webp_to_avif % ]\e[0m"

# Rename output JPG/Webp with ssim, filesize, and quality/parameters used
mv out.jpg  "CVVDP-${jpgcvvdp} ${jpgsize}KB [ $jpgbase $jpgq ].jpg"
mv out.webp "CVVDP-${webpcvvdp}_${webpsize}KB [ $webpbas $webpq ].webp"
mv out.jxl "CVVDP-${jxlcvvdp}_${jxlsize}KB [ $jxlbase $jxlq ].jxl"
mv out.avif "CVVDP-${avifcvvdp}_${avifsize}KB [ $avifbas $avifq ].avif"

# nuke everything or quit
echo "" && read -p "Press enter to nuke files" exit
rm -f jpg.png webp.png png.ppm jxl.png avif.png
rm -f "CVVDP-${jpgcvvdp} ${jpgsize}KB [ $jpgbase $jpgq ].jpg"
rm -f "CVVDP-${webpcvvdp}_${webpsize}KB [ $webpbas $webpq ].webp"
rm -f "CVVDP-${jxlcvvdp}_${jxlsize}KB [ $jxlbase $jxlq ].jxl"
rm -f "CVVDP-${avifcvvdp}_${avifsize}KB [ $avifbas $avifq ].avif"

2026-07-14·16:28·No.947·

>>951

Good job with the script, man 👍
Very nice bird too

2026-07-15·03:57·No.951

JPG_flicker.webp

>>947
Thanks. I forgot to mention that in
>>944 I used the windows CVVDP implementation so if you use this you have to make it point to your /mnt/c/Users/X directory of CVVDP on windows not your linux cvvdp-env. WSL1 is pretty cool since now you can use BASH scripts but it seems to have performance problems with the python programs. I was about to give up and go back to FCVVDP but it's cool that microsoft allows you to call native windows programs even from a BASH script.

I've been thinking about what kind of flicker test to do for this thread and I've settled on webp vs jpg. The 9.5 CVVDP one shows roughly the same overall degradation except on outlines of the beak/feathers where some mosquito/ringing noise seems to radiate out a little. The 9.0 CVVDP shows even more unwanted distortion on outlines. To be fair I am using an outdated JPG encoder and making it extra hard by using full 4:4:4 but still pretty useful info IMHO.

Other thing I've decided on is I'll catbox the 9.0 and 9.5 files per image. Though I'm not sure if I should paste 12 separate catbox links on my planned reddit thread or jam them into one giant catbox link.

2026-07-15·03:59·No.952

Here's the modified flicker test bash script for in.png vs *.webp vs *.jpg. There must be a better way to make this thing adaptive but this works so I won't mess with this for now.

#!/usr/bin/env bash

for f in in.png; do
ffmpeg -hide_banner -loglevel error -i "$f" \
-vf "format=rgb24,drawtext=text='$(basename "$f")':fontcolor=white:fontsize=12:box=1:[email protected]:boxborderw=8:x=20:y=20" \
-pix_fmt rgb24 png.png
done

for f in *.webp; do
ffmpeg -hide_banner -loglevel error -i "$f" \
-vf "format=rgb24,drawtext=text='$(basename "$f")':fontcolor=white:fontsize=12:box=1:[email protected]:boxborderw=8:x=(w-text_w)/2:y=20" \
-pix_fmt rgb24 webp.png
done

for f in *.jpg; do
ffmpeg -hide_banner -loglevel error -i "$f" \
-vf "format=rgb24,drawtext=text='$(basename "$f")':fontcolor=white:fontsize=12:box=1:[email protected]:boxborderw=8:x=w-text_w-20:y=20" \
-pix_fmt rgb24 jpg.png
done

img2webp -lossless -m 6 -loop 0 -d 1000 png.png webp.png jpg.png -o JPG_flicker.webp
rm -f png.png webp.png jpg.png

2026-07-15·04:08·No.953·

>>955

JPG_flicker2.webp

Here is the 9.0 CVVDP flicker comparison. Looking at it more carefully, it's a mixed bag depending differing subjective opinions. The outlines are preserved much better in Webp but some fine detail in the feathers is destroyed. Still I think both would still be considered medium fidelity here, neither one is terrible or great IMHO.

Also the catbox for the 9.0 and 9.5 files: https://files.catbox.moe/srfamu.zip

2026-07-15·04:22·No.954

CVVDP-9.53_51.7KB [ -m 6 -af -q 81 ].webp

Also I guess the CVVDP 9.5 Webp result is worth keeping here as well. I'll also include the YUV PSNR generated with cwebp

Saving file 'out.webp'
File:      in.png
Dimension: 960 x 769
Output:    52942 bytes Y-U-V-All-PSNR 42.31 45.50 45.81   43.16 dB
           (0.57 bpp)

Webp CVVDP= 9.53 Fsize= 51.7 KB [ cwebp -m 6 -af -q 81 ]

My guess is if after all these 12 images the dB value doesn't vary too much, maybe that could point to a "safe" upper range. Though it would also have to be combined with the Anime image result as well. That or just keep track of what the higher cwebp -quality value was used. Dunno which one might be more practical/useful.

2026-07-15·12:36·No.955·

>>956

>>953
This animated webp is working great for these comparisons, but I noticed the website is leaving that retract icon on the top right and overlapping the text. Next website update I'm gonna remove that

2026-07-16·05:21·No.956

IMG_2.webp

>>955
Sounds good but maybe leave it on the thumbnail. I kinda like knowing that I'm looking at a thumbnail and not the full size image.

2nd image codec comparison is finally done. This is the first time I've seen Webp reach parity with JXL on a photographic image in high fidelity. Even stranger is Webp achieved greater compression efficiency at medium fidelity. I'm not sure what makes this image so cursed for JXL.

Image source for picrel being tested: https://en.wikipedia.org/wiki/File:ALP-45DP_4504_On_Head_of_Train_1009,_at_Hoboken_Terminal_6-1-12.jpg

catbox for 9.0 and 9.5 files: https://files.catbox.moe/jbx94e.zip

2026-07-19·00:50·No.962

9.5-CVVDP_Webp-vs-JPG_flicker.webp.webp

2nd image

9.5 CVVDP flicker comparison (Webp vs JPG)

2026-07-19·00:52·No.963·

>>964

9.0-CVVDP_Webp-vs-JPG_flicker.webp

2nd image

9.0 CVVDP flicker comparison (Webp vs JPG)

2026-07-19·04:17·No.964·

>>968

>>963
I found this one very interesting, JPG looks horrible, but webp also had some issues with details missing like parts of the wipers

2026-07-20·03:31·No.965

IMG_3.webp

3rd image

Image sauce: https://en.wikipedia.org/wiki/File:Fr%C3%BChling_bl%C3%BChender_Kirschenbaum.jpg

9.0 and 9.5 files: https://files.catbox.moe/ujxh75.zip

2026-07-20·03:36·No.966

9.5-CVVDP_Webp-vs-JPG_flicker.webp

3rd image

9.5 CVVDP flicker comparison (Webp vs JPG)

2026-07-20·03:41·No.967

9.0-CVVDP_Webp-vs-JPG_flicker.webp

3rd image

9.0 CVVDP flicker comparison (Webp vs JPG)

2026-07-20·03:50·No.968

>>964
I see it now. Strange how selective lossy compression can be depending on the image codec. I guess Webp didn't think preserving the part was important. Maybe this is what the photo preset addresses but I won't test this for now to keep the test consistent.

Dam now I can't unsee that, a floating wiper blade. A wireless wiper blade lol.

2026-07-20·05:19·No.969

IMG_4.webp

4th image

Image sauce: https://en.wikipedia.org/wiki/File:ANA_B737-700%28JA02AN%29_%22Gold_Jet%22_%283943310662%29.jpg

9.0 and 9.5 files: https://files.catbox.moe/rjm53f.zip

2026-07-20·05:22·No.970

9.5-CVVDP_Webp-vs-JPG_flicker.webp

4th image

9.5 CVVDP flicker comparison (Webp vs JPG)

2026-07-20·05:25·No.971·

>>987

9.0-CVVDP_Webp-vs-JPG_flicker.webp

4th image

9.0 CVVDP flicker comparison (Webp vs JPG)

2026-07-21·00:01·No.987

>>971
This one is interesting because even though the lines on the ground don't disappear, they change colors lol

[Catalog]

[Auto]

21/12

LOSSY JPG/Webp/JXL/AVIF

CVVDP parameters and info:

JPG encoding parameters and version:

cjpeg -baseline -sample 1x1,1x1,1x1 -quality X in.ppm > out.jpg
Independent JPEG Group's CJPEG, version 6b 27-Mar-1998

Webp encoding parameters and version:

cwebp -m 6 -af -q X in.png -o out.webp
libwebp v1.6.0 (released on June 30, 2025)

JXL encoding parameters and version:

cjxl in.png out.jxl -e 10 -q X
libjxl v0.12.0 (released on July 1, 2026)

AVIF encoding parameters and version:

#!/bin/bash jpgbase="-baseline -sample 1x1,1x1,1x1 -quality" && read -p "jpg quality: " jpgq webpbas="-m 6 -af -q" && read -p "cwebp quality: " webpq jxlbase="-e 10 -q" && read -p "jxl quality: " jxlq avifbas="-y 420 -d 10 -s 0 -q" && read -p "avifenc quality: " avifq && echo "" CVVDPWIN="/mnt/c/Users/1337/AppData/Local/Programs/Python/Python313/Scripts/cvvdp.exe" # JPG enc/dec ffmpeg -hide_banner -loglevel quiet -i in.png -y png.ppm && cjpeg $jpgbase $jpgq png.ppm > out.jpg ffmpeg -hide_banner -loglevel quiet -i out.jpg -y jpg.png # Webp, JXL, AVIF enc/dec cwebp $webpbas $webpq in.png -o out.webp 2>&1 | head -n 5 && dwebp -quiet out.webp -o webp.png cjxl in.png out.jxl $jxlbase $jxlq --quiet && djxl out.jxl jxl.png --quiet avifenc $avifbas $avifq in.png out.avif &>/dev/null && avifdec out.avif avif.png &>/dev/null

# Display CVVDP, filesize, and parameters for JPG jpgcvvdp=$($CVVDPWIN -t jpg.png -r in.png --device cpu -d standard_fhd -m cvvdp 2>&1 \ | awk -F'=' '/cvvdp=/{gsub(/ \[JOD\]/,"",$2); printf "%.2f",$2}') jpgsize=$(awk 'BEGIN{printf "%.1f",'$(stat -c%s out.jpg)'/1024}') && echo "" echo "JPG CVVDP= $jpgcvvdp Fsize= $jpgsize KB [ cjpeg $jpgbase $jpgq ]" && echo "" # Display CVVDP, filesize, and parameters for Webp webpcvvdp=$($CVVDPWIN -t webp.png -r in.png --device cpu -d standard_fhd -m cvvdp 2>&1 \ | awk -F'=' '/cvvdp=/{gsub(/ \[JOD\]/,"",$2); printf "%.2f",$2}') webpsize=$(awk 'BEGIN{printf "%.1f",'$(stat -c%s out.webp)'/1024}') echo "Webp CVVDP= $webpcvvdp Fsize= $webpsize KB [ cwebp $webpbas $webpq ]" && echo "" # Display CVVDP, filesize, and parameters for JXL jxlcvvdp=$($CVVDPWIN -t jxl.png -r in.png --device cpu -d standard_fhd -m cvvdp 2>&1 \ | awk -F'=' '/cvvdp=/{gsub(/ \[JOD\]/,"",$2); printf "%.2f",$2}') jxlsize=$(awk 'BEGIN{printf "%.1f",'$(stat -c%s out.jxl)'/1024}') echo "JXL CVVDP= $jxlcvvdp Fsize= $jxlsize KB [ cjxl $jxlbase $jxlq ]" && echo "" # Display CVVDP, filesize, and parameters for AVIF avifcvvdp=$($CVVDPWIN -t avif.png -r in.png --device cpu -d standard_fhd -m cvvdp 2>&1 \ | awk -F'=' '/cvvdp=/{gsub(/ \[JOD\]/,"",$2); printf "%.2f",$2}') avifsize=$(awk 'BEGIN{printf "%.1f",'$(stat -c%s out.avif)'/1024}') echo "AVIF CVVDP= $avifcvvdp Fsize= $avifsize KB [ avifenc $avifbas $avifq]" && echo ""

# Diminishing returns comparison webp_to_avif=$(awk "BEGIN {printf \"%.1f\", (($jpgsize-$webpsize)/($jpgsize-$avifsize))*100}") webp_to_jxl=$(awk "BEGIN {printf \"%.1f\", (($jpgsize-$webpsize)/($jpgsize-$jxlsize))*100}") echo "" && echo -e "\e[32mWebP compression efficiency relative to [ JXL: $webp_to_jxl % ] & [ AVIF: $webp_to_avif % ]\e[0m" # Rename output JPG/Webp with ssim, filesize, and quality/parameters used mv out.jpg "CVVDP-${jpgcvvdp} ${jpgsize}KB [ $jpgbase $jpgq ].jpg" mv out.webp "CVVDP-${webpcvvdp}_${webpsize}KB [ $webpbas $webpq ].webp" mv out.jxl "CVVDP-${jxlcvvdp}_${jxlsize}KB [ $jxlbase $jxlq ].jxl" mv out.avif "CVVDP-${avifcvvdp}_${avifsize}KB [ $avifbas $avifq ].avif" # nuke everything or quit echo "" && read -p "Press enter to nuke files" exit rm -f jpg.png webp.png png.ppm jxl.png avif.png rm -f "CVVDP-${jpgcvvdp} ${jpgsize}KB [ $jpgbase $jpgq ].jpg" rm -f "CVVDP-${webpcvvdp}_${webpsize}KB [ $webpbas $webpq ].webp" rm -f "CVVDP-${jxlcvvdp}_${jxlsize}KB [ $jxlbase $jxlq ].jxl" rm -f "CVVDP-${avifcvvdp}_${avifsize}KB [ $avifbas $avifq ].avif"

#!/usr/bin/env bash for f in in.png; do ffmpeg -hide_banner -loglevel error -i "$f" \ -vf "format=rgb24,drawtext=text='$(basename "$f")':fontcolor=white:fontsize=12:box=1:[email protected]:boxborderw=8:x=20:y=20" \ -pix_fmt rgb24 png.png done for f in *.webp; do ffmpeg -hide_banner -loglevel error -i "$f" \ -vf "format=rgb24,drawtext=text='$(basename "$f")':fontcolor=white:fontsize=12:box=1:[email protected]:boxborderw=8:x=(w-text_w)/2:y=20" \ -pix_fmt rgb24 webp.png done for f in *.jpg; do ffmpeg -hide_banner -loglevel error -i "$f" \ -vf "format=rgb24,drawtext=text='$(basename "$f")':fontcolor=white:fontsize=12:box=1:[email protected]:boxborderw=8:x=w-text_w-20:y=20" \ -pix_fmt rgb24 jpg.png done img2webp -lossless -m 6 -loop 0 -d 1000 png.png webp.png jpg.png -o JPG_flicker.webp rm -f png.png webp.png jpg.png

Saving file 'out.webp' File: in.png Dimension: 960 x 769 Output: 52942 bytes Y-U-V-All-PSNR 42.31 45.50 45.81 43.16 dB (0.57 bpp) Webp CVVDP= 9.53 Fsize= 51.7 KB [ cwebp -m 6 -af -q 81 ]