<h1 id="magful-krea-2-thick-paint-style">magful — Krea 2 Thick-Paint Style</h1><p>A thick-paint / impasto style LoRA for Krea 2 with a built-in colour strategy — not just a texture filter. It doesn't lock your composition and doesn't override your art style.</p><p>Trigger word: magful — OPTIONAL. The LoRA is always-on; adding it makes no systematic difference.</p><p>Recommended: strength 1.0 | cfg 1.0 (daily) or 1.5 (punchier)</p><h2 id="read-first-want-an-empty-scene-scenery-only">⚠️ READ FIRST — WANT AN EMPTY SCENE / SCENERY ONLY?</h2><h3 id="this-is-where-people-get-tripped-up.-it-absolutely-can-do-empty-scenes-short-prompts-just-can't-hold-back-its-default-configuration.">This is where people get tripped up. It absolutely can do empty scenes — short prompts just can't hold back its default configuration.</h3><p>38 empty-scene renders (6 scene types × 3 seeds):</p><p>- Short prompt (~10 words) + poster-art prefix: characters appeared 18/18</p><p>- Short + "no humans" at the very front: 14/14 (even with no extra words at all)</p><p>- Medium detail (~47 words): 0/2 — clean</p><p>- Long prompt (200+ words, per-object spec): 0/2 — clean</p><p>- Still-life subject (no poster prefix): 0/6 — clean</p><p>- "dark fantasy aesthetic" tone: 0/4 — clean</p><h2 id="what-to-do">WHAT TO DO</h2><h3 id="1)-write-enough-objects-and-space-counts-positions-materials-light-direction.">1) Write enough objects and space — counts, positions, materials, light direction.</h3><p>Same empty classroom: 10 words → 4-7 students fill the desks; 47 words (blackboard, window direction, dust, chair states) → completely empty; 234 words (five rows of thirty desks, clock at 4:10, aluminium window frames, long floor shadows) → still empty, and composition adherence is excellent.</p><h3 id="2)-avoid-the-poster-art-prefix.">2) Avoid the poster-art prefix.</h3><h3 id="&quot;official-poster-art-cinematic-composition&quot;-semantically-means-&quot;a-key-visual-with-a-subject&quot;.-same-empty-classroom-only-the-prefix-changed:">"official poster art / cinematic composition" semantically means "a key visual WITH a subject". Same empty classroom, only the prefix changed:</h3><p>- no prefix / magful / "no humans" first: classroom still gets filled with students</p><p>- "Modern high-end anime illustration, official poster art, ..." : even a still life becomes a silver-haired portrait</p><p>- "dark fantasy aesthetic": classroom stays completely empty</p><p>Moving "no humans" to the front does NOT fix it (4/4 either way).</p><p>3) Use cfg 1.5 — cuts the head-count a lot (5-6 people → 1), but doesn't remove it.</p><p>One line: empty scenes come from writing detail, not from adding negatives.</p><h2 id="what-it-does">WHAT IT DOES</h2><p>Controlled experiment (fixed prompt × 4 seeds × 3 strengths, 36 renders):</p><p>LoRA colorfulness = 0.896 × base colorfulness + 3.93 (R² = 0.995, strength 1.0)</p><p>Crossover at base ≈ 37.7 — below it the LoRA ADDS colour, above it MUTES.</p><p>- Gothic / dark (nun in cathedral): base 15.1 → 18.5 (×1.23, more colour)</p><p>- Mid-tone (afternoon street): base 35.8 → 34.5 (×0.96, about the same)</p><p>- Saturated (magenta/cyan flower field): base 95.8 → 69.9 (×0.73, muted)</p><p>In both directions: distinguishable colours +40% to +174%, luminance micro-detail +3% to +49%, stronger colour peaks (the most saturated 5%), especially in dark scenes.</p><p>It doesn't simply "add saturation" or "go grey" — it pulls every starting point toward a "composed but powerful" middle range. Low-chroma images gain colour, high-saturation images gain gradation. Both make the picture read thicker.</p><h2 id="strength-and-settings">STRENGTH &amp; SETTINGS</h2><p>Strength 1.0 — thickest layering, most restrained chroma (recommended)</p><p>Strength 0.8 — middle ground</p><p>Strength 0.6 — flattening out</p><p>Strength 0.4 — close to base</p><p>cfg: 1.0 for daily use, 1.5 for more punch (chroma +5.6% to +21.6%). The colour behaviour keeps the same direction at both.</p><p>Sampling: euler / sgm_uniform / 8 steps | CLIP: Qwen3-VL 4B | VAE: qwen_image_vae</p><p>CONTROLLABILITY (all backed by A/B tests)</p><p>- Explicit hair colour: 13/13 (black / blonde / green / pink / brown / silver / purple)</p><p>- Explicit head-count on plain background: 6/6 exact</p><p>- Explicit photographic style: 8/8 fully photoreal — the LoRA steps aside</p><p>- Six forced media (ink line art / oil / 3D / photo / watercolour / flat vector): 6/6 kept their medium — NOT invasive</p><p>Invasiveness measured: with ink line art forced, luminance gradient 0.1152 vs base 0.1154 — a difference of 0.0002. It only nudges inside a medium (mutes vector colour, thickens watercolour strokes). It will not destroy the style you asked for.</p><p>Prompt responsiveness: across 6 very different prompts, within-group distance 0.31 (LoRA) vs 0.36 (base) — distinguishing ability preserved, only ~13% homogenisation.</p><h1 id="things-to-watch">THINGS TO WATCH</h1><p>- Unspecified hair tends to silver (the base model already prefers this)</p><p>- Unspecified outfit tends to school uniform (same-seed test: 6/6 uniforms vs 6/6 casual)</p><p>- Scenes implying company may add a second figure — add "solo, only one person in frame"</p><p>- Short prompt + poster prefix fills the frame with people — see the first section</p><h1 id="retracted-claims">RETRACTED CLAIMS</h1><h3 id="several-earlier-conclusions-were-overturned-by-later-controlled-tests-three-of-them-from-this-description's-own-earlier-drafts.-listed-so-nobody-propagates-them:">Several earlier conclusions were overturned by later controlled tests — three of them from this description's own earlier drafts. Listed so nobody propagates them:</h3><p>- "Overrides explicit photographic style" → 8/8 stayed photoreal, the LoRA stepped aside</p><p>- "Miscounts people in group scenes" → 6/6 exact with "3girls" on a plain background</p><p>- "Obeys empty-scene prompts 4/4" → that test prompt read "solo, 1girl, only one person in frame, a tiny figure" — it asked for a person. Empty scenes were never tested then.</p><p>- "Has a colorfulness floor around 28-31" → dark scenes output 15.6-20.8, no floor; the real relation is linear shrinkage</p><p>- "+91% colorfulness in dark scenes" → single uncontrolled sample; controlled value is +10% to +20%</p><p>- "Lighting words are bound to characters, so they pull people in" → the no-word control also produced people 14/14; the deciding factor is prompt specificity</p><p>"Silver hair" is NOT this LoRA's signature — the base model draws silver-haired girls from "1girl" alone.</p><h1 id="example-prompts">EXAMPLE PROMPTS</h1><h3 id="reusable-prefix:">Reusable prefix:</h3><p>magful, Modern high-end anime illustration, official poster art, cinematic composition, intricate details, sophisticated style, depth of field, sharp focus on subject, masterpiece.</p><p>1. 1girl, gothic nun, silver cross, gothic cathedral, heavy rain, dark dramatic lighting</p><p>2. 1girl, valkyrie, silver winged helmet, ornate plate armor, holding a spear, divine golden light, clouds and sky, epic battlefield, dramatic backlighting</p><p>3. 1girl, wandering swordswoman, black hair in a high ponytail, worn traveler cloak, katana at her side, cherry blossom grove, wind scattering petals, twilight</p><p>4. 1girl, young librarian, brown hair with glasses, cream blouse, standing among towering bookshelves, warm lamplight, dust motes, cozy atmosphere</p><p>5. 1girl, elf archer, long green hair, pointed ears, dark green cloak with gold embroidery, drawing a wooden bow, ancient mossy forest, fireflies, moonlight</p><p>6. 1girl, cosmic priestess, white hair, layered white and gold robe, standing before a giant celestial harp, starfield, floating planets, ornate golden architecture</p><h3 id="negative:">Negative:</h3><p>cartoon, chibi, cel shading, flat colors, childlike, photorealistic, 3d render, low quality, bad anatomy, text, watermark</p><h2 id="training">TRAINING</h2><p>Base: Krea 2 | trainer: ai-toolkit | 2500 steps / 138 epochs | rank 32</p><p>Dataset: a self-curated set (mostly own AI-generated works, plus a small number of third-party AI-generated pieces), style-consistent but structurally diverse; the dataset was not retained after training.</p><p>License: non-commercial. Because the dataset includes third-party material, please do not use commercially; derivatives should credit the source.</p><p></p><p>---</p><h3 id="mirrors"><strong>Mirrors</strong></h3><p>- Hugging Face: <a target="_blank" rel="ugc" href="https://huggingface.co/weiermod/magful-krea2-lora">https://huggingface.co/weiermod/magful-krea2-lora</a></p><p>- GitHub: <a target="_blank" rel="ugc" href="https://github.com/weiernmod/magful-krea2-lora">https://github.com/weiernmod/magful-krea2-lora</a></p><p></p><h2>中文说明</h2><p>magful —— Krea 2 厚涂风格 LoRA。触发词 magful 非必须，默认生效。推荐强度 1.0，cfg 1.0（日常）/ 1.5（更艳）。</p><p>想做空景？关键在提示词的具体程度，不在题材。38 张空景测试：短句（10 词）+ 海报前缀 → 18/18 出人；加到 47 词 → 0/2；234 词 → 0/2；静物题材 → 0/6；dark fantasy aesthetic 调性 → 0/4。official poster art 这类前缀是主要开关，连静物都会被画成银发女孩，把 no humans 放最前面救不回来。空景靠「写详细」，不靠「加负面」。</p><p>效果：LoRA 鲜艳度 = 0.896 × 基座鲜艳度 + 3.93（R²=0.995），交叉点约 37.7 —— 低于它加色（暗调 ×1.23），高于它减色（高饱和 ×0.73）。两个方向都提升色阶（+40~174%）和明暗微细节（+3~49%）。</p><p>可控性：发色 13/13、白底人数 6/6、摄影风格 8/8 让位、六种强制画风 6/6 全部守住（墨线风格下梯度只差 0.0002）。</p><p>注意：不写发色会银白、不写服装会制服、场景暗示有同伴会多一个人。</p><p>许可：非商用。素材含第三方 AI 作品，训练后未保留。</p><h1></h1><p></p>

Files with SHA256: d9bb332a73dee01eae65246504c47c73a43b58982f804e61a4c149bca6c5a22d