<p>A style LoRA for <strong>Krea 2</strong> that brings the Midjourney photoreal look — soft film colors, warm side light, deep shadows, background blur and the typical "Midjourney face" — to portraits of Japanese women. Trained on 48 Midjourney images of adult women in their 20s and 30s.</p><h3 id="trigger">Trigger</h3><p><code>mjjpg</code> — start your prompt with <code>mjjpg.</code></p><h3 id="recommended-settings-(tested)">Recommended settings (tested)</h3><ul><li><p>Model: Krea 2 Turbo (fp8), text encoder qwen3vl_4b, VAE qwen_image_vae</p></li><li><p>8 steps, CFG 1, euler / simple, empty negative (ConditioningZeroOut)</p></li><li><p>LoRA strength: <strong>0.8–1.0</strong> (0.8 and 1.0 look almost the same; 0.7 for a lighter touch)</p></li><li><p>Size: 832×1248 (2:3) works well</p></li></ul><h3 id="camera-color-presets">Camera / color presets</h3><p>Add one of these sentences at the end of the prompt, exactly as written. Leave it out for the default look. The effect is a <strong>subtle color shift</strong>, not a different style.</p><ul><li><p><code>Shot on Contax T2 with Fujifilm Natura 1600, Black Mist 1/4 filter.</code> — compact film camera, a bit more contrast; great with direct flash</p></li><li><p><code>Shot on Leica M6 with Kodak Portra 400.</code> — slightly warmer film snapshot</p></li><li><p><code>Shot on Canon EOS R5 with RF 50mm F1.2 L.</code> — clean digital, a little more background blur</p></li><li><p><code>Shot on Fujifilm Pro 400H overexposed by two stops.</code> — meant to be airy and pale; currently the weakest preset</p></li></ul><h3 id="other-options-learned-from-captions">Other options learned from captions</h3><ul><li><p>Light eyes: <code>pale grey eyes</code> (tends to make the face look more mixed/Western — drop it or lower the strength to keep a Japanese look)</p></li><li><p>Body: <code>slender build with a small bust and narrow hips</code> / <code>average build with a medium bust</code> / <code>curvy figure with a full bust and wide hips</code></p></li><li><p>Age: <code>in her 20s</code> / <code>in her 30s</code></p></li><li><p>Framing: <code>close-up portrait</code> / <code>upper body shot</code> / <code>thigh-up shot</code> / <code>full body shot</code>, and <code>from the front</code> / <code>in three-quarter view</code> / <code>in side profile</code> / <code>from behind, looking back over her shoulder</code></p></li><li><p>Makeup: <code>red lipstick</code> / <code>smoky eye makeup</code> / <code>glossy pink lips</code></p></li><li><p>Light: write the direction, e.g. <code>with soft light from the left</code>, <code>lit by direct flash</code>, <code>with backlight from the setting sun</code></p></li></ul><h3 id="prompt-template">Prompt template</h3><p><code>mjjpg. A full body shot from the front of a Japanese woman in her 20s with a long black braid. She wears a white linen dress and brown sandals, standing with both feet on the sand, looking at the camera with a soft smile. She stands on a beach at sunset, with backlight from the setting sun.</code></p><p>Captions were written as natural English in this order: trigger → framing and person → clothes, pose and expression → place and light direction → camera sentence. Prompts in the same order work best. Every showcase image includes its full prompt.</p><h3 id="notes">Notes</h3><ul><li><p>The style works even with very short prompts (e.g. <code>mjjpg. a Japanese woman in a cafe</code>); the presets and options need the exact wording above.</p></li><li><p>Full body shots work — mention shoes and the floor/ground (<code>standing with both feet on the floor</code>).</p></li><li><p>Text on signs comes out garbled (base model limitation).</p></li><li><p>All training images show adults. Not intended for NSFW use.</p></li></ul><p></p>

Files with SHA256: 3ced5f0fed83d47077f95244f982f3b4cbe0c58222d8d766410a7b1081bfb04a