<h1 id="uncanima">Uncanima</h1>Trigger: <code>unc4nim4</code>Uncanima is my first custom style LoRA for Anima, trained as an experiment in building a flexible semi-real 3D digital illustration style. The look sits somewhere between stylized digital painting, polished 3D render, and uncanny almost-real glamour. It is not meant to be pure photography, but with the right prompting it can get surprisingly close to realistic.Let's be real. It's a total gooner model. It handles spicy glamour, pinup framing, erotic posing, lingerie, exposed skin, outdoor NSFW, and moody cinematic portraits very well. That said, it is not locked to NSFW. I did train it on a few male oriented, landscape and even abstract images from my custom dataset. <h2 id="style-behavior">Style Behavior</h2>Uncanima changes a lot depending on how you prompt it.Simple subject-only prompts tend to produce a softer semi-real digital painting look:<code>unc4nim4, 1girl, solo, portrait, long black hair, leather jacket</code>More detailed prompts with lighting, lens, composition, and setting can push it much closer to realistic:<code>unc4nim4, 1girl, solo, close-up portrait, long black hair, black leather jacket, neon city street, night lighting, shallow depth of field, cinematic crop, soft rim light</code>So if the output looks too painterly, add lighting, composition, environment, and camera language. If it looks too realistic, too glossy, or too intense, simplify the prompt and lower the LoRA strength.<h2 id="recommended-strength">Recommended Strength</h2>LoRA strength depends heavily on the model stack and what other LoRAs you are using.Most of my preview images were generated using this LoRA alongside a turbo LoRA at around 0.7 strength.Suggested starting points:<ul><li>0.5–0.7 for subtle style influence or when stacking with other LoRAs</li><li>0.7–0.9 for stronger Uncanima style</li><li>1.0 for full style force, especially on simple prompts</li><li>Lower the strength if it starts overpowering character LoRAs or making everything too glossy</li></ul><h2 id="prompting-tips">Prompting Tips</h2>Use the trigger near the front:<code>unc4nim4, ...</code>The model responds well to visible, concrete prompt details:<ul><li>lighting: <code>golden hour</code>, <code>soft rim light</code>, <code>neon reflections</code>, <code>warm lamplight</code></li><li>composition: <code>close-up portrait</code>, <code>low angle</code>, <code>looking back over shoulder</code>, <code>centered composition</code></li><li>setting: <code>sunny beach</code>, <code>dark bedroom</code>, <code>rainy city street</code>, <code>snowy cabin</code>, <code>greenhouse</code></li><li>texture/material: <code>wet pavement</code>, <code>silk sheets</code>, <code>glossy leather</code>, <code>lace lingerie</code>, <code>fur blanket</code></li><li>body/pose detail: <code>kneeling</code>, <code>bent over</code>, <code>seated sideways</code>, <code>one knee raised</code>, <code>arms behind head</code></li></ul>For NSFW images, be specific about what is visible and what the clothing is doing. The model responds better to direct clothing interaction than vague “sexy pose” prompting.<h2 id="example-prompt">Example Prompt</h2><code>masterpiece, best quality, amazing quality, very aesthetic, unc4nim4, explicit, 1girl, solo, adult woman, kneeling on a beach towel, bikini top untied, bare breasts, long wet brunette hair, soft curvy body, white sand beach, turquoise water, bright summer sun, shallow depth of field</code><h2 id="notes">Notes</h2>This LoRA was trained as a style experiment, not a character LoRA. It is meant to bring a flexible Uncanima look to whatever subject you prompt. Because the training set includes a lot of glamour and NSFW material, it may naturally lean spicy unless prompted otherwise.For SFW work, use clear safe subjects, clothing, and setting details. For stronger realism, add lighting and composition. For softer digital painting, keep the prompt simpler.Have fun. The model is shiny, strange, flexible, and very easy to push in different directions.

Files with SHA256: f42e3978739bc1fc2693ab2daf9310daeb964f1cdd88b4c4e9bae6c320830d52