Why I Wrote a Pixel-by-Pixel Multiply Blend in PHP-GD to Beat Dompdf

#wordpress #php #webdev #pdf

A real estate broker came to me with a familiar problem. Every property exposé — the glossy multi-page PDF that goes to potential buyers — was being assembled by hand in an Adobe layout tool. Address typed in. Prices typed in. Photos imported one by one. Floor plans appended. Logo placed. Disclaimer pasted. PDF exported. Filed somewhere. Sent.

That's 45 minutes per listing. And the moment a price changes, you're starting over.

The agency was already on WordPress. JetEngine was already managing properties as custom post types. So the path was clear: generate the exposé directly from the data — one click, one PDF, pixel-perfect corporate design.

The first 80% of that plugin was straightforward. PHP, Dompdf, an HTML template, FPDI to merge in the floor plan PDFs. Done in a week.

The last 20% almost killed me.

The Dompdf opacity problem

The cover page of an exposé is one of those classic real-estate looks: a hero photo of the property, with a dark gradient overlay so the headline is legible.

In CSS, that's three lines:

.cover {
    background-image: url(...);
    background-blend-mode: multiply;
}
.overlay {
    opacity: 0.6;
    background: linear-gradient(...);
}

In Dompdf, both background-blend-mode and opacity are unreliable. They render inconsistently between Dompdf versions, between fonts, between paper sizes. I spent three hours coaxing different combinations before accepting that this isn't going to work.

The obvious move would be to swap libraries. mPDF? wkhtmltopdf? Browserless?

But all of those add operational weight: wkhtmltopdf is a binary that needs to be installed and updated on the host. Browserless is an external service. mPDF would mean rewriting the templating layer.

The agency wanted something they could deploy on standard WordPress hosting. No binaries. No external services. No black boxes.

So I went the other way.

Pre-compute the blend in PHP-GD

If Dompdf can't blend at render time, the blend has to happen before Dompdf sees the image. Which means doing it in PHP, with GD, pixel by pixel.

// Multiply-blend two images of the same size
function multiply_blend($base, $overlay) {
    $width  = imagesx($base);
    $height = imagesy($base);

    for ($y = 0; $y < $height; $y++) {
        for ($x = 0; $x < $width; $x++) {
            $b = imagecolorat($base, $x, $y);
            $o = imagecolorat($overlay, $x, $y);

            // Multiply formula per channel: (a × b) / 255
            $r = (($b >> 16 & 0xFF) * ($o >> 16 & 0xFF)) / 255;
            $g = (($b >> 8  & 0xFF) * ($o >> 8  & 0xFF)) / 255;
            $bl = (($b      & 0xFF) * ($o      & 0xFF)) / 255;

            $color = imagecolorallocate($base, $r, $g, $bl);
            imagesetpixel($base, $x, $y, $color);
        }
    }
    return $base;
}

That's brute-force math, twice the resolution times two channels per pixel. On a 1600×900 cover image, that's roughly 1.4 million iterations per blend. Slow.

Then you encode the result as JPEG with high quality and embed it as a base64 data-URI directly in the HTML that goes into Dompdf:

ob_start();
imagejpeg($blended, null, 92);
$encoded = base64_encode(ob_get_clean());
$dataUri = "data:image/jpeg;base64,{$encoded}";

Dompdf doesn't know it's looking at a blended image. It just sees <img src="data:image/jpeg;..."> and embeds the bytes. Pixel-perfect output.

The cache makes it fast

Brute-force blending is fine if you don't do it often. The plugin generates each PDF once, then caches it in wp-content/uploads/. It only re-renders when the underlying data changes — and it knows that via a hash:

function inex_get_or_generate_pdf($post_id) {
    $cached_version  = get_post_meta($post_id, '_inex_pdf_cache_version', true);
    $current_version = inex_get_cache_version($post_id);

    if (file_exists($filepath) && hash_equals($cached_version, $current_version)) {
        return $fileurl; // Serve from cache, sub-second
    }

    if (!inex_acquire_pdf_generation_lock($post_id)) {
        return new WP_Error('locked', 'PDF being generated, please retry');
    }

    try {
        $data = inex_collect_expose_data($post_id);
        inex_render_expose_pdf_file($data, $filepath . '.tmp');
        rename($filepath . '.tmp', $filepath); // atomic
        update_post_meta($post_id, '_inex_pdf_cache_version', $current_version);
    } finally {
        inex_release_pdf_generation_lock($post_id);
    }
}

Three things worth flagging in this snippet:

hash_equals for the cache check — not strictly security-critical here, but it sets the right habit. Same function that protects token comparison.
rename on a .tmp file is atomic on POSIX filesystems. A concurrent reader never sees a half-written PDF.
try/finally for the lock release. If rendering throws, the lock still clears. No "stuck PDFs" after a failed deploy.

The cache version is computed by hashing every relevant meta field plus the modification times of the gallery images. Change the price → hash changes → next download regenerates. Nobody pushes a "refresh" button. Nobody runs a cron job.

When does this approach make sense?

A custom PHP-GD blend plus Dompdf isn't elegant. It's a workaround stacked on a workaround. Six hundred lines of plugin code do the work that one CSS rule would have done if Dompdf supported it.

But it ships in a single PHP plugin. No binaries. No external dependencies. No vendor lock-in. Standard WordPress hosting handles it. Cold render: 5–15 seconds. Cached read: under one second.

The lesson, for me, was that "swap the library" isn't always the right reflex. Sometimes the cheapest path forward is a precomputation stage in your own stack that hides the limitation from the layer that has it.

You can read the full case study (German) here: hafenpixel.de/hinter-den-kulissen/expose-generator-jetengine-wordpress

Or the English version: hafenpixel.de/en/behind-the-scenes/expose-generator-jetengine-wordpress