mrjschulte
commented
7 months ago Do you have an example image?
Open the-database opened 7 months ago
mrjschulte
commented
7 months ago Do you have an example image?
the-database
commented
7 months ago Here is the image I was using, zipped to make sure the image file is untouched by github.
If I load this image into chaiNNer and then save it to another file and then compare the files, the new file appears to be less saturated than the original file when I load them into Photoshop to compare.
Here's the output from magick identify -verbose .\GAvGRBhaUAApWr3.jpg which shows the embedded ICC profile:
Image:
Filename: .\GAvGRBhaUAApWr3.jpg
Permissions: rw-rw-rw-
Format: JPEG (Joint Photographic Experts Group JFIF format)
Mime type: image/jpeg
Class: DirectClass
Geometry: 816x508+0+0
Resolution: 72x72
Print size: 11.3333x7.05556
Units: Undefined
Colorspace: sRGB
Type: TrueColor
Base type: Undefined
Endianness: Undefined
Depth: 8-bit
Channels: 3.0
Channel depth:
Red: 8-bit
Green: 8-bit
Blue: 8-bit
Channel statistics:
Pixels: 414528
Red:
min: 0 (0)
max: 255 (1)
mean: 194.361 (0.762202)
median: 185 (0.72549)
standard deviation: 34.7114 (0.136123)
kurtosis: 1.81024
skewness: -0.271293
entropy: 0.674755
Green:
min: 0 (0)
max: 255 (1)
mean: 154.072 (0.604204)
median: 116 (0.454902)
standard deviation: 57.8123 (0.226715)
kurtosis: -1.16799
skewness: 0.620514
entropy: 0.689868
Blue:
min: 0 (0)
max: 255 (1)
mean: 155.619 (0.61027)
median: 118 (0.462745)
standard deviation: 58.8426 (0.230755)
kurtosis: -1.27659
skewness: 0.531043
entropy: 0.713966
Image statistics:
Overall:
min: 0 (0)
max: 255 (1)
mean: 168.017 (0.658892)
median: 139.667 (0.547712)
standard deviation: 50.4554 (0.197864)
kurtosis: -0.211444
skewness: 0.293421
entropy: 0.692863
Rendering intent: Perceptual
Gamma: 0.454545
Chromaticity:
red primary: (0.64,0.33,0.03)
green primary: (0.3,0.6,0.1)
blue primary: (0.15,0.06,0.79)
white point: (0.3127,0.329,0.3583)
Matte color: grey74
Background color: white
Border color: srgb(223,223,223)
Transparent color: black
Interlace: JPEG
Intensity: Undefined
Compose: Over
Page geometry: 816x508+0+0
Dispose: Undefined
Iterations: 0
Compression: JPEG
Quality: 85
Orientation: Undefined
Profiles:
Profile-icc: 536 bytes
Properties:
date:create: 2024-02-20T14:21:20+00:00
date:modify: 2024-02-20T14:21:20+00:00
date:timestamp: 2024-02-21T06:18:47+00:00
icc:copyright: Copyright Apple Inc., 2022
icc:description: Display P3
jpeg:colorspace: 2
jpeg:sampling-factor: 2x2,1x1,1x1
signature: d23f47ac61796d02028f846d321329d88e4941fa215a11025870888dc76c830a
Artifacts:
verbose: true
Tainted: False
Filesize: 59412B
Number pixels: 414528
Pixel cache type: Memory
Pixels per second: 79.0014MP
Time-to-live: 0:0:0:0 2024-02-21T06:18:47Z
User time: 0.005u
Elapsed time: 0:01.005
Version: ImageMagick 7.1.1-25 Q16-HDRI x64 98e7513:20231230 https://imagemagick.orgCould you maybe attach screenshots of the versions of the image you have in various apps that clearly show the difference between what other apps are showing (assuming the stored icc profile is being honored and applied) and what you see in chaiNNer?
Also, for a basic support preflight here, which OS are you on and what version of chaiNNer are you using?
Motivation When processing images with embedded ICC profiles and saving the output, the colors in the saved output don't look as expected compared to the colors of the input image. Based on the discussion on Discord, chaiNNer currently does not support color profiles and reads in all images as sRGB.
Description This is a complex feature and there are many ways it could be designed and implemented. As long as there is a way to handle images with embedded ICC profiles, so that exported images don't unexpectedly shift colors, any solution works for me.
Alternatives Converting an image with embedded ICC profile to sRGB in another software (such as Photoshop) allowed me to use the sRGB image in chaiNNer without any issues. However, many people wouldn't know that this workaround is necessary for handling images with ICC profiles. For example, when I first encountered this when upscaling a specific image, I thought the color shift was simply a flaw of the model, but it turned out that wasn't the case.