<h3 id="ranni-the-witch-or-modular-4-in-1-character-lora">Ranni The Witch | Modular 4-in-1 Character LoRA</h3>1. RECOMMENDED SETTINGS (Important!)<ul><li>Base Model: waiIllustriousSDXL_v110 (Highly recommended) or base Illustrious-XL.</li><li>LoRA Weight (waiIllustrious): 0.7 - 1.0 (Check the grid examples).</li><li>LoRA Weight (Base Illustrious): 0.5 - 0.8.</li><li>CFG Scale: 2.5 - 5.5 (Lower CFG helps stabilize the 4-arm anatomy).</li><li>Sampler: DPM++ 2M SDE Heun / DPM++ 2M Karras.</li></ul>2. BASE SKELETON &amp; CORE TRIGGERSStart your generation with this baseline.<ul><li>Mandatory Anchor: ryomen_core (Must be in ALL prompts. Anchors the 4 arms, spectral face, and closed/glowing eye. Contains NO clothing).</li><li>Base Positive: masterpiece, best quality, very aesthetic, absurdres, 1girl, solo, ryomen_core, [Insert Style &amp; Camera Tags], &lt;lora:Ranni_EldenWai:0.8&gt;</li><li>Base Negative: lowres, text, error, missing, extra, fewer, cropped, jpeg artifacts, worst quality, bad quality, watermark, unfinished, chromatic aberration, human skin, modern clothes, two_arms, two_hands</li></ul>3. BODY FORMS Use these tags to control her physical state (especially for nude concepts).<ul><li>ranni_doll — Forces the puppet texture (rope connections). Note: The spectral face may fade, as the model interprets it as a soulless shell.</li><li>ranni_waifu — Forces a porcelain/human texture with cracks, removing the rope joints.</li><li>Note: If generating her fully clothed, these tags are optional.</li></ul>4. WARDROBE CONTROLClothing is completely modular.<ul><li>Canonical Outfit: witch_hat, fur_cloak, white_dress</li><li>No Cloak: Remove fur_cloak from the positive prompt and add fur_cloak, outerwear to the Negative Prompt.</li></ul>5. TROUBLESHOOTING &amp; NEGATIVE PROMPTING<ul><li>Missing Arms: Add two_arms, two_hands to the Negative Prompt.</li><li>Bleeding Joints (in Waifu mode): Add doll_joints to the Negative Prompt.</li><li>Eye Control (Overrides): The ryomen_core tag defaults to one closed/glowing eye. If the base model forces both eyes open on complex angles, add closed eye or one eye closed to your Positive prompt, and open eyes to your Negative prompt. You can use both simultaneously for a hard override.</li><li>Form Conflicts: Cross-negate the forms (e.g., put ranni_doll in negative when using ranni_waifu).</li><li>Stubborn Poses: For highly complex geometry (like lying on the ground), use img2img with a basic sketch.</li></ul>6. INSTALLATION<ol><li>Download the .safetensors file.</li><li>Place the file into your WebUI LoRA folder. For standard A1111 or WebUI Forge, the path is: models/Lora</li><li>Refresh your LoRA list in the UI, click on the model to add it to your prompt, and ensure you include the mandatory trigger word ryomen_core.</li></ol><h3 id="prompt-engineering-cheat-sheet">🛠️ PROMPT ENGINEERING CHEAT SHEET</h3>I trained this LoRA using a strict tagging matrix. Use these exact tags from the dataset to get the most stable results:Render Styles &amp; Medium (Pick 1-2):<ul><li>2D/Art: digital_art, anime_art, detailed_illustration, painterly, impasto, heavy brush strokes, watercolor (style), pastel (style), textured</li><li>Manga/Graphic: monochrome, greyscale, spot_color, comic, manga, screentones, lineart</li><li>3D/Dark Fantasy: 3d, cg, realistic, video_game, dark_fantasy</li></ul>Camera Angles &amp; Framing:<ul><li>Shot Size: close-up, hand_focus, medium_close-up, medium_shot, cowboy_shot, full_body</li><li>Horizontal View: front_view, profile, from_side, three-quarter_view, from_behind</li><li>Vertical View: from_above, high_angle, from_below, low_angle</li><li>Focus: looking_at_viewer, looking_away, looking_up, looking_down</li></ul>Poses, Hands &amp; Kinematics:<ul><li>Body: standing, sitting, sitting_on_rock, lying, squatting, crossed_legs</li><li>Arms/Hands: arms_crossed, arms_outstretched, hands_clasped, hands_together, hands_on_lap, steepled_fingers, hand_to_face, hand_on_hip</li><li>Actions: casting_spell, open_hands, reaching_up, holding_staff, holding_book</li></ul>Lighting, Atmosphere &amp; Depth:<ul><li>Lighting: dramatic_lighting, high_contrast, chiaroscuro, diffusing light, soft lighting, sharp rim lighting</li><li>Atmosphere: fog, haze, dust, glowing_motes, tyndall effect, sunbeams</li><li>Lens/Depth: depth of field, blurry background, blurry foreground, atmospheric perspective, bokeh, dutch angle</li></ul>P.S. A little tech-flex: I ran an extreme stress test by merging this Ranni model with my upcoming 4-armed Heian Sukuna LoRA. Yes, the success rate is 1 in 8, and it requires a dozen crutches to render properly. Casuals will see a pile of noise, but engineers will understand the absolute miracle of the engine processing conflicting multi-arm geometry without instantly crashing into a fleshy mess. The Sukuna LoRA is coming next.<h2 id="p.p.s.-if-you-want-to-support-my-caffeine-addiction-or-request-custom-b2b-lora-training-all-relevant-links-will-be-available-on-my-profile-soon.">P.P.S. If you want to support my caffeine addiction or request custom B2B LoRA training, all relevant links will be available on my profile soon. </h2>

Files with SHA256: feeb49bd8f46bbe92e00d8ff84429c280533c7f3685a4a6c4d64662acc654d11