Composite blend parameters like clear, source not working as expected.

[g-abhishek]

What are you trying to achieve?
I am trying to have blend transformation, which is one of the property of composite api, for my image transformation and stucked in a output which is different than given in the cairographics.org website

Have you searched for similar questions?
Yes, but didn't got any solution

Are you able to provide a minimal, standalone code sample that demonstrates this question?

try {
    await sharp("largeImage.jpeg")
      .composite([
        {
          input: "small.jpeg",
          gravity: “center”,
          blend: “clear”, // source, dest-out, etc
        },
      ])
      .toFile(“output.jpeg”);
  } catch (error) {
    console.log(error);
  }
}

Are you able to provide a sample image that helps explain the question?
Input images :-

Output images :-
for blend type clear,

for blend type source,

[lovell]

Hi, most compositing blends/operators require an alpha channel on at least one image. The two input images provided here are both JPEGs, which does not support an alpha channel.

[g-abhishek]

I tried with PNGs images, with one image(big one) has alpha channel, but still getting the same output.

Input Images :-

Output Images :-
for blend type clear,

for blend type source,

[lovell]

Those outputs appear to be correct for clear and source, which aren't particularly useful blend modes. Perhaps you intended to use over (the default)?

[g-abhishek]

If you see the https://www.cairographics.org/operators/ documentation, output given here is different than the generated one for clear and souce

[g-abhishek]

@lovell Can you please refer the https://www.cairographics.org/operators/ docs once and check is this correct or not?

[lovell]

The way the composite operation works has been improved via commit c620025

I noticed that when compositing 3 or more images, the accuracy of the blend mode could suffer, and the clear and source modes particularly so as they would consider the combination of the previous 2 layers rather than 1st layer. Thanks for bringing this to my attention - let's track this at #2286