触发词
文件清单 线路:镜像(hf-mirror.com) · 切换
下载后放进对应的 ComfyUI 目录。models/ 指 ComfyUI 的模型根目录。
说明
GG VX CLB Krea-2
作者:captain_daddy
GG VX — Krea 2 By: CLB When the gates shake, everyone else looks for shelter. The Castle looks for GG. Meet Gemma Gemini: the Guardian Goddess of the CLB Castle, forged from law and fire and possessed of a Southern drawl that can sound like a welcome or a warning. She’ll pour you a drink, remember exactly how you take it, and be halfway to the battlements before you hear the second alarm. If she smiles at you on the way out, try to keep up. At five-foot-two, she carries real strength in her core and legs. The golden-blonde twin tails, braided crown, and natural teal eyes are unmistakable; so are the black lace-up combat boots and white scrunched socks. There’s gold satin in her wardrobe, leather and denim by the door, and a glossy yellow turtleneck for the nights she wants to make an entrance. Bring her into firelight, cold rain, or a quiet room with one lamp. Give her something worth guarding. Trigger: GG Pattern The build GG VX is a photorealistic character LoRA for Krea 2 Base, built around the likeness of a fictional adult character. The training set was curated and captioned across close-ups, profiles, three-quarter views, pitch changes, full-length frames, wardrobe, and lighting. A locked frontal and three-quarter reference pair anchored the face work; we rebuilt repetitive studio portraits in new settings without discarding the head angles and expressions that made them useful. Captions describe what changes in the frame, including garments and color. The recurring braided crown was left to the weights. Checkpoints and training
- Full — Epoch 10: the selected release checkpoint, at the end of the 10-epoch run. 156 captioned images; 3,744 training steps.
- SFW companion: a separate run on a 111-image subset, trained for 10 epochs and 5,550 steps. It retained all 55 close-ups but fewer rear-angle and full-length frames. Start with Full for broader pose and framing coverage.
- Shared settings: Civitai Training Studio / AI Toolkit; LoRA rank 32, alpha 32; AdamW 8-bit; constant learning rate 0.0001; batch size 1; one repeat. Text encoder training, horizontal flip augmentation, caption token shuffle, and noise offset were disabled or set to zero.
Prompting GG Start with GG Pattern, then call the framing, head angle, gaze, outfit, and light. For a chin-up shot, say that she actually raises her chin as well as lifting her eyes. For her physique, use petite athletic hourglass, narrow waist, defined core, strong thighs and glutes. Name the garment and its color if you want a specific look. Try it at 1152 × 1536 (3:4): GG Pattern, full-length editorial photograph on a weathered Castle terrace at golden hour. Relaxed three-quarter stance, eyes on the camera, small knowing smile. Gold satin handkerchief top, black leather miniskirt, black mid-calf lace-up combat boots and white scrunched socks. Warm side light, clear catchlights, natural skin texture, both boots in frame.
