<h1 id="real-blonde"><span style="color:rgb(34, 139, 230)">Real Blonde</span></h1><p>Apparently most models never seen natural blondes. They usually default to dyed hair with dark roots and dark eyebrows. And since I didn't find any attempt, I made LoRAs for real(istic) blondes.</p><p>Training data created with Krea2, SDXL, Chroma: Straight blonde hair (including braids and bangs), natural light to darker eyebrows, bright eyecolors, pale or tanned skin. Strong focus on upper body.</p><h1 id="blondies-forever"><span style="color:rgb(250, 176, 5)">Blondies Forever ❤️</span></h1><p>I decided to create two LoRAs - one with naturally light eyebrows (which can seem almost invisible - <strong>realblonde1</strong>), and one leaning into the more mainstream dyed look (<strong>realblonde2</strong>).</p><p>I feel it grew more into an eyebrow LoRA ;)</p><p></p><p>Somehow I forgot to upload the Krea2 versions, so here we go.</p><h1 id="anima:"><span style="color:rgb(34, 139, 230)">Anima:</span></h1><p>0.6-0.9. Works great on Turbo and Base. Plays well with Anima styles - see below.</p><pre><code>realblonde, natural blonde woman, blonde eyebrows</code></pre><h1 id="chroma:"><span style="color:rgb(34, 139, 230)">Chroma:</span></h1><p>Chroma is the actual reason I wanted to train this LoRA. It is insanely bad at blondes - and so good at everything else.</p><p>Don't use "realblonde" or you will end up with shirt prints or corner tags.</p><h1 id="zit:"><span style="color:rgb(34, 139, 230)">ZiT:</span></h1><p>The Z-Image version is straight forward: Use the triggers and start from 0.7.</p>

Real Blonde

<p>This is a <strong>subset </strong>of the training data, two versions: 8 and 10 epochs.</p>

Files with SHA256: f739057c1bc280578fcd711dcaa75bbdd6f31dd5a32a46062c3653d16c63a634