This is a LoRa I decided to train on gpt-img-2. There's two goals: improve facial texture and increase body proportions. Ideogram by default can't move away from a slim body. This pushes it to be nearly as good as the 200 elo better gpt-img. It happened to be mainly Korean, I generated most of the images before Ideogram was even released.<h2 id="trigger">Trigger</h2>Use:<code>Korean</code>Place it once near the start of the woman’s description:<code>Korean, woman with...</code>Because <code>Korean</code> is also a normal word understood by the base model, it may affect facial appearance and styling as well as the trained body shape. Tags That Work Well<ul><li><code>large/full projecting chest</code> — 288 images, 93%</li><li><code>narrow shoulders</code> — 236, 76%</li><li><code>narrow or sharply indented waist</code> — 214, 69%</li><li><code>compact short-waisted torso</code> — 209, 68%</li><li><code>pronounced underbust taper</code> — 197, 64%</li><li><code>broad or rounded hips</code> — 163, 53%</li></ul>Appearance and clothing<ul><li><code>dark or black hair</code> — 294, 95%</li><li><code>long dark hair</code> — about 177, 57%</li><li><code>white clothing</code> — 287, 93%</li><li><code>fitted or taut clothing</code> — 275, 89%</li><li><code>fitted white shirt/blouse/top</code> — 262, 85%</li><li><code>high-waisted</code> — 234, 76%</li><li><code>mini skirt</code> — 226, 73%</li><li><code>button-down blouse</code> — 202, 65%</li><li><code>gray clothing</code> — 201, 65%</li><li><code>pleated skirt</code> — 196, 63%</li><li><code>gray pleated skirt</code> — 141, 46%</li><li><code>black sheer tights</code> — 128, 41%</li><li><code>black skirt</code> — 86, 28%</li><li><code>white crop top</code> — about 62, 20%</li></ul>Settings<ul><li><code>pool</code>, <code>flooded interior</code>, or visible water — 106, 34%</li><li><code>night</code> — 72, 23%</li><li><code>apartment</code>, bedroom, or simple indoor room — 71, 23%</li><li><code>street</code>, sidewalk, or crosswalk — 60, 19%</li><li><code>flooded corridor</code> or flooded room — 58, 19%</li><li><code>indoor pool</code> — 48, 16%</li><li><code>wet clothing</code> — 42, 14%</li><li><code>alley</code>, passage, or corridor — 35, 11%</li><li><code>wet pavement</code> and urban reflections — 26, 8%</li><li><code>instant-photo border</code> — 21, 7%</li></ul>Poses and camera<ul><li><code>three-quarter stance</code> — 121, 39%</li><li><code>high-angle</code> or <code>overhead view</code> — 100, 32%</li><li><code>front-facing</code> or <code>near-front stance</code> — 82, 27%</li><li><code>selfie perspective</code> — 27, 9%</li><li><code>side profile</code> — 27, 9%</li><li><code>low-angle view</code> — 13, 4%</li></ul><h3 id="default-dataset-look">Default dataset look</h3><code>Korean, woman with long dark hair, a compact short-waisted torso, a large projecting chest, sharply indented waist, and broad rounded hips, wearing a fitted white short-sleeve blouse tucked into a gray high-waisted pleated mini skirt and black sheer tights, three-quarter stance.</code><h3 id="water-theme">Water theme</h3><code>Korean, woman wearing a fitted wet white blouse and gray pleated mini skirt, standing waist-deep in an indoor pool or flooded corridor, cool fluorescent lighting, dark reflective water, three-quarter view.</code><h3 id="night-street-theme">Night street theme</h3><code>Korean, woman with long dark hair wearing a fitted white blouse and high-waisted dark mini skirt, standing beside a crosswalk on a wet city street at night, reflective pavement, soft storefront lighting, smartphone photograph.</code><h3 id="bright-indoor-theme">Bright indoor theme</h3><code>Korean, woman in a fitted white crop top and high-waisted gray or beige mini skirt, standing in a bright apartment near a large window, soft daylight, high-angle smartphone view.</code><h3 id="uniform-like-theme">Uniform-like theme</h3><code>Korean, woman with long black hair wearing a fitted white short-sleeve button blouse, dark neck ribbon, gray pleated mini skirt, and black sheer tights, near-front three-quarter pose.</code><h2 id="compact-tag-string">Compact Tag String</h2><code>Korean, long dark hair, fitted white blouse, gray high-waisted pleated mini skirt, black sheer tights, compact short-waisted torso, large projecting chest, narrow waist, broad rounded hips, three-quarter stance, high-angle smartphone photograph</code>For the strongest in-distribution result, combine the morphology with white fitted clothing, a gray pleated mini skirt, and either a pool/flooded interior or a wet nighttime street.<h2 id="recommended-strength">Recommended strength</h2>Start around:<ul><li><code>0.6-0.8</code></li></ul>The LoRA can influence body proportions even without the trigger, so lower the strength if unrelated woman prompts change too much.<h2 id="basic-promptkorean-woman-with-a-compact-short-waisted-torso-a-full-projecting-upper-torso-relative-to-a-narrow-ribcage-a-sharply-defined-waist-and-rounded-hips.-wearing-a-fitted-blouse-tucked-into-a-high-waisted-skirt-standing-in-a-three-quarter-pose.">Basic prompt <code>Korean, woman with a compact short-waisted torso, a full projecting upper torso relative to a narrow ribcage, a sharply defined waist, and rounded hips. Wearing a fitted blouse tucked into a high-waisted skirt, standing in a three-quarter pose.</code></h2><h2 id="stronger-morphology">Stronger morphology</h2>Use more direct geometry:<code>Korean, woman with a compact short-waisted torso, massive breasts, deep cleavage and projection, strong upper-torso-to-waist contrast, a sharply indented natural waist, and broad rounded hips.</code><h2 id="important-note">Important note</h2>The LoRA does not act as a perfect on/off trigger.Tests showed:<ul><li>loading the LoRA can change body proportions even without <code>Korean</code></li><li>adding <code>Korean</code> may also shift the face toward a Korean appearance</li><li>explicit body descriptions still matter</li></ul>For the strongest control, describe the body shape you want instead of relying only on the trigger.<h2 id="example-structured-prompt">Example structured prompt</h2><pre><code>{"aspect_ratio":"5:8","high_level_description":"A natural low-light smartphone photograph of a woman standing in a narrow rain-wet city alley at night, framed from above her hair to the upper thighs.","compositional_deconstruction":{"background":"A narrow brick service alley with closed shutters, exposed cables, small practical lamps, and wet pavement reflecting subdued cyan and amber light.","elements":[{"type":"obj","bbox":[20,70,1000,910],"desc":"Korean, woman with a compact short-waisted torso, very full projecting breasts relative to a narrow ribcage, a sharply indented natural waist, and rounded hips. Her breasts are extremely large. She wears a fitted white button-down blouse pulled taut across the upper torso and tucked into a dark red high-waisted pencil skirt. Near-front three-quarter stance, upright posture, both arms behind her back, long dark hair, and a calm direct gaze."}]}}
</code></pre><h1 id="how-it-was-trained">How It Was Trained</h1><ul><li>Base model: Ideogram 4 FP8</li><li>Training tool: ostris AI Toolkit</li><li>LoRA rank/alpha: 32/32</li><li>Training steps: 2500 but this one is 1750</li><li>Optimizer: AdamW 8-bit</li><li>Learning rate: <code>1e-4</code></li><li>Resolutions: 512, 768, and 1024</li><li>Text encoder: not trained</li><li>Trigger word: <code>Korean</code></li><li>Dataset: 309 retained image-caption json. Captioned manually by GPT-5.5 High with some guidance.</li><li>Captions used structured descriptions of body geometry, clothing tension, pose, camera perspective, and scene content. Mostly softened, not actually saying "huge breasts".</li></ul>The dataset contains entirely synthetic Korean women from gpt-img-2, so the trigger may influence both morphology and facial appearance.

Korean Body enhancer | Ideogram

Initial release. First training run. Great results but the dataset wasn't generated intending to train a model.

Files with SHA256: f5b7ee5145e0e11f7a526b60e999517a108e8ee03d18a628d953cabaa0014735