Taash V1

詳細

ファイルをダウンロード (1)

このバージョンについて

Version 2 Notes

Dataset Feedback: Retrained using a "best-of" selection from v1.1 generations to reinforce the character's unique facial structure and piercing eye detail.
SDXL 1.0 Native: Formally trained on the SDXL 1.0 base model for maximum compatibility with modern realistic checkpoints.
Improved Consistency: Reduced the "generic AI" look by doubling down on the original character's edgy, athletic aesthetic.
Anatomy Refinement: While still a work in progress, this version was tuned to be more stable with limb placement compared to the initial release.
Texture Polish: Enhanced skin rendering for a more natural, photographic texture.

Suggested Prompts & Settings

Base Prompt Template:

score_9, score_8_up, score_7_up, Taash, a 20-year-old woman, athletic build, edgy aesthetic, piercing eyes, realistic skin texture, [insert outfit/action], cinematic lighting, depth of field, 8k resolution

Negative Prompt (Crucial for SDXL):

(fused limbs:1.2), (extra limbs:1.2), (merged limbs:1.2), (malformed hands:1.2), (connected limbs:1.2), extra fingers, ugly, deformed, mutated, gross proportions, long neck, missing arms, missing legs.

Technical Recommendations:

Sampling Method: DPM++ 2M Karras or Euler a
Sampling Steps: 25 - 35
CFG Scale: 5 - 7
LoRA Weight: 0.8 is the sweet spot.
- Use 0.6 if you want more of the Base Model's style.
- Use 1.0 if you want her specific face to be extremely dominant.

モデル説明

Model Description: Taash V1 (Original Character)

Overview Taash is an original character (OC) developed through a specialized AI-to-AI pipeline. The concept originated in ChatGPT and was refined through the Civitai generator. This LoRA features a custom-tuned dataset that includes manual photographic references for her eyes to ensure a unique and piercing look.

Character Profile

Age: 20 years old.
Build: Slim, athletic, and toned.
Aesthetic: Realistic and edgy. While she fits well into modern or street-style renders, she is versatile enough to be placed in fantasy or "darker" settings.
Themes: This model is capable of generating NSFW content and maintains consistent character features across different outfits.

Usage Recommendations

Trigger Word: Taash
LoRA Weight: 0.7 - 0.85 (Start at 0.8).
Negative Prompting: Essential for stability (see below).

Known Quirks & Troubleshooting

Important Note on Anatomy: > This model is currently prone to limb merging and body deformities (extra limbs or fused arms/legs), especially in wide-angle shots or complex poses.

How to get the best results:

Negative Prompting: Use these keywords to help keep her limbs separate:
- fused limbs, extra limbs, malformed hands, long neck, morphed limbs, extra fingers, connected limbs
Framing: Performs most consistently with portrait or upper-body shots.
Refining: I highly recommend using ADetailer for faces and hands to clean up any structural artifacts during the generation process.

Recommended Negative Prompt

To minimize the "limb merging" quirk, use this string in your negative prompt box:

このモデルで生成された画像

並び替え

モデルタイプ	LORA
ベースモデル	SDXL 1.0
公開日	2026-03-13
トレーニングワード	taash