CivArchive
    Ass sitting from behind (Krea2) - v1.0
    NSFW
    Preview 138664542
    Preview 138664568
    Preview 138664643

    This LoRA is designed to emphasize and frame the buttocks of female subjects (18+) in a seated position, with the camera specifically focusing on the glutes from a rear viewpoint.

    The model highlights the seated silhouette with a stylized, dramatic flair, while maintaining a suggestive and erotic aesthetic with minimal genitalia reveal. It is specialized for rear‑angle seated poses.


    Subject & Pose Description

    The subject is depicted sitting, with the buttocks exposed and the camera framing them from behind. The model responds best to natural language prompts that describe the seated posture and the rear focus.

    For greater variety in poses, it is recommended to combine this LoRA with other LoRA models that introduce different body angles, environments, or stylistic elements.


    Strength

    0.5- 1.0: Subtle enhancement and framing of the seated glutes
    1.1-2.2: Stronger, more pronounced stylized effect


    Image Format Recommendation

    Most images in the training dataset were processed in vertical orientation.

    Therefore, it is highly recommended to generate images in portrait mode (9:16 aspect ratio) for optimal results.


    Training & Captioning Methodology

    • Each training image underwent detailed analysis and captioning, taking approximately 4 minutes per image across 8 distinct sections.

    • Captions frequently exceed 300 words to capture every nuance of the composition, including object and body descriptions, environment, and camera framing.


    Trigger Words & Concepts

    While the LoRA responds strongly to natural language prompts describing the pose and focus, it works best with concepts such as:

    • sitting

    • exposed buttocks

    • from behind

    • rear view

    • seated pose

    • ass focus

    Best Practice

    This model was trained exclusively on natural language captions generated by a vision model (Qwen). No Danbooru-style tags were used in the training data.

    For best results, use descriptive full sentences in your prompts rather than comma-separated tag lists.


    Combining with Other LoRAs

    To achieve a wider variety of poses and stylistic variations, it is strongly recommended to mix this LoRA with other LoRA models. This can introduce different body angles, clothing styles, environments, or artistic interpretations while retaining the core seated‑rear focus.


    License & Usage

    • You are free to use this model in your projects, including merging or mixing it with other models.

    • If you include this LoRA in any merge or mix, please give proper credit.


    Support the Creator

    If you enjoy my work and want to help cover the electricity costs required to train and refine these models, please consider supporting me:

    Address: https://ko-fi.com/weirddreamer (Donations are always appreciated!)

    Description

    Dataset Size:

    116 high-quality, manually curated images

    Data Processing:

    Images were meticulously filtered using advanced duplicate removal tools to ensure maximum variety and uniqueness.

    Captions:

    Tagging was performed using Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-Q4_K_M

    FAQ

    LORA
    Krea 2

    Details

    Downloads
    594
    Platform
    CivitAI
    Platform Status
    Available
    Created
    8/3/2026
    Updated
    8/12/2026
    Deleted
    -

    Files

    WeirdDreamer_ass-sitting_from_behind-krea2.safetensors