Remove Background but Keep Glows, Halos, and Soft Edges

You've created the perfect image - a glowing neon sign, a magical character with a luminous aura, or a glass object catching the light. Now you need it on a transparent background. You run it through a background remover and... the glow is gone. The soft edges are jagged. The magic is lost.

This is the fundamental limitation of every background removal tool. They weren't built to handle semi-transparent pixels. But there's an alternative that preserves every last bit of transparency data.

The Problem With Background Removal

Background removal tools - remove.bg, Photoshop's "Remove Background," Canva's background eraser, and every AI-powered cutout tool - all work the same way. They analyze your image and classify every pixel as either "subject" or "background."

This is a binary decision. Each pixel is either kept at 100% opacity or discarded entirely. There is no middle ground.

For subjects with hard, well-defined edges - a product on a white table, a person standing in front of a wall - this works fine. The boundary between subject and background is clear.

But the moment your image has semi-transparent elements, this binary approach falls apart. A glow that should fade gradually from 80% opacity to 0% gets sliced at some arbitrary threshold. Pixels above the threshold become fully opaque. Pixels below it vanish. The smooth gradient becomes a hard cliff.

What Gets Destroyed

Semi-transparent elements are more common than you might think. Here's what background removal tools consistently damage or destroy:

• Soft glows - neon text, magical effects, light sources that radiate outward with decreasing opacity
• Halos and rim lighting - the bright outline around backlit subjects that fades into transparency
• Smoke, fog, and atmospheric effects - wispy elements that are inherently semi-transparent
• Glass and translucent materials - objects that partially show what's behind them
• Gradient fades and feathered edges - any edge that transitions gradually rather than abruptly
• Light rays and bloom effects - beams of light that overlap with the background
• Color fringing at edges - traces of the old background color that bleed into edge pixels

All of these share the same characteristic: they are partially transparent. They don't fully belong to the subject or the background - they exist in between, blending with whatever is behind them. That's what makes them look natural, and that's exactly what background removers can't handle.

Alpha Recovery: The Alternative

Instead of deciding which pixels to keep and which to discard, there's a technique that calculates the exact transparency value of every pixel. It's called alpha channel recovery.

The method uses two versions of the same image: one rendered on a white background and one on a black background.

The math behind it is elegant:

• Pixels that look identical on both backgrounds are fully opaque - the background doesn't show through at all
• Pixels that are completely different (white on the white version, black on the black version) are fully transparent - they're 100% background
• Everything in between has a precise, calculable transparency value based on how much the pixel changed between the two versions

This means a pixel that's 30% transparent stays exactly 30% transparent. A glow that fades from full brightness to invisible retains every step of that gradient. No clipping, no hard edges, no lost data.

The result is a mathematically perfect alpha channel that preserves 100% of the glow, halo, and soft edge information from your original image.

This technique was originally documented by jidefr for use with Nano Banana, but it works with any image source where you can produce white and black background versions.

How to Preserve Glows Step by Step

Step 1 - Generate or render on white

Create your image with a pure white background:

• AI generators: Include "pure white background" or "on a white background" in your prompt
• 3D software: Set the background/environment color to white (#FFFFFF) before rendering
• Compositing: Place your subject on a solid white layer

The key is that the background must be pure white - not off-white, not light gray. Any deviation will affect the accuracy of the transparency calculation.

Step 2 - Generate or render on black

Create the exact same image with a pure black background. The subject must be identical - only the background color changes.

• AI generators: Use the same prompt with "pure black background" instead. If your tool supports seeds, use the same seed value
• 3D software: Change only the background color to black (#000000) and re-render with identical settings
• Compositing: Swap the background layer from white to black

Keeping everything else identical is critical. Same seed, same settings, same prompt - only the background color should differ.

Step 3 - Recover with Transparify

Upload both versions to transparify.app. The algorithm compares every pixel between the two images and calculates the exact transparency value.

• Use the before/after comparison slider to verify the result - drag it to see the original vs the recovered transparency
• If you accidentally swapped the white and black images, use the swap button to fix them
• Export as PNG for lossless quality, or WebP for smaller file sizes with an adjustable quality slider

Everything runs entirely in your browser. Your images are never uploaded to any server.

Background Removal vs Alpha Recovery - Side by Side

When to Use Each Approach

Use alpha recovery when:

• Your image has glows, soft edges, or light effects that need to be preserved
• The subject includes smoke, glass, fog, or other translucent elements
• You need mathematically perfect transparency - no guessing, no approximation
• You can generate or render two versions of the image (white and black backgrounds)
• You're working with AI-generated images from ChatGPT, Midjourney, Gemini, Flux, or similar tools

Use background removal when:

• The subject has hard, well-defined edges - product photos, people, objects with clear outlines
• You only have one image and can't produce a second version on a different background
• Speed matters more than quality - you need a quick cutout and minor edge artifacts are acceptable
• The background is complex or textured - a real-world photo where you can't control the background color