Initial VTF support

[REDxEYE]

This pull request adds initial support for VTF image format

Changes proposed in this pull request:

• Add support for Valve Texture Format images

[REDxEYE]

Can I provide arguments to save the function? Like compression mode, flags and etc?

[radarhere]

Like im.save("out.vtf", compression=...)? Yes. Those arguments will become part of the im.encoderinfo dictionary.

[REDxEYE]

Yes, that's just what I needed, thanks!

[REDxEYE]

@radarhere Do I need to encode the image row-by-row or i can go like 4 rows at once because Bcn works in blocks of 4 by 4 pixels?
I never worked with Pillow C API, so i need some help

[radarhere]

@radarhere Do I need to encode the image row-by-row or i can go like 4 rows at once because Bcn works in blocks of 4 by 4 pixels? I never worked with Pillow C API, so i need some help

No, you don't have to.

A while ago, I attempted to implement DXT1 encoding. You might find that helpful - radarhere@363fab4

[REDxEYE]

I see, your attempt is quite close to what I wrote, but I used a different way to get 2 main colors for the block. Can I reuse some of your code from 363fab4?

[radarhere]

Feel free

[REDxEYE]

@radarhere Can you please review my bcn encoder?

[radarhere]

Slightly confused - this PR is still a draft, but you'd like a review?

Also, your lint problems could be resolved by just using the black formatting - REDxEYE#2. Is there a particular part of the styling that you dislike?

[REDxEYE]

Slightly confused - this PR is still a draft, but you'd like a review?

Also, your lint problems could be resolved by just using the black formatting - REDxEYE#2. Is there a particular part of the styling that you dislike?

I'm asking for a encoder code review, pr itself not even close to be ready :)

[radarhere]

#8807 has been merged with something close to my previous encoder.

I'm not opposed to using a different approach to improve it, but in the tests I added, I have

out = str(tmp_path / "temp.dds")
with Image.open(TEST_FILE_DXT1) as im:
  im.convert("RGB").save(out, pixel_format="DXT1")
assert_image_similar_tofile(im, out, 1.84)

If I run that against your encoder, I get https://github.com/radarhere/Pillow/actions/runs/13935788461/job/39003253301

AssertionError: average pixel value difference 4.1112 > epsilon 1.8400
assert 1.84 >= 4.1111907958984375

If I save the outputs, comparison.zip, and zoom in to visually compare them,

I think it is clear that the version in main is closer to the original test file than the version in this PR.

Feel free to disagree.

[REDxEYE]

Ye, will update PR later

[radarhere]

Looking at https://developer.valvesoftware.com/wiki/VTF_(Valve_Texture_Format), it says

DXT1_ONEBITALPHA format does not properly work; use regular DXT1 with 1-bit alpha flag enabled instead.

Do you mind if I remove support from this PR?

[REDxEYE]

Ye, i'm totally okay

[radarhere]

What does 'etalon' refer to?

[REDxEYE]

supposed to be like reference image or something :)

[radarhere]

All of the test images you've included are able to be distributed as part of Pillow under our license, yes?

[REDxEYE]

Yes, these are test images i created myself, so they are non-copyrighted