v4 Highlights: Specifically optimized for (naked miniguy) to fix the "stubborn clothing" issue found in previous versions.
Description:
This is a LoRA model focused on Giantess concepts, integrating elements of Vore, Unbirth, and Miniguy interactions.
Recommended Strength: 0.8
Usage Tip: At 0.8 strength, it is highly compatible with other character LoRAs.
Core Keywords:
Base Trigger: giantess
Vore Scenes: vore
Unbirth Scenes: unbirth
Interactions/Stepping: miniguy
Training Focus:
The training dataset focuses exclusively on conceptual actions. Since a significant portion of the material involves close-up shots (particularly near the vaginal area), these scenarios may appear more frequently during random generation.
Troubleshooting Size Issues:
If the miniguy appears too large during stepping or interaction scenes, try increasing the prompt weight to (miniguy:1.4) to improve the scale, or run multiple generations (Random Seeds) to find the best result.
"Scale is dynamic in Anima 2B. If the size isn't right, try descriptive comparisons like 'smaller than her toe' and roll a few more seeds. The AI's imagination is part of the fun!"
The above is a Google machine translation
下方中文
v4 更新重點： 針對 (naked miniguy) 進行專門優化，徹底解決前幾代模型中「衣服難以脫除」的頑疾。
這是一個giantess概念要素的Lora模型，融合了 vore unbirth 和 miniguy 互動，
建議Strength:0.8
Strength:0.8 可搭配Lora人物使用
通用關鍵字是 giantess
吞噬時 vore
放入陰道時 unbirth
單獨與小人互動例如踩踏時 miniguy
訓練素材只針概念動作去做訓練，因為素材比較陰道附近的場景,所以一直Random Seeds會比較常出現
踩踏如果小人尺寸不夠小，建議將拉高權重到(miniguy:1.4)可能會稍微改善，或多Random Seeds幾張圖
（在 Anima 2B 中比例是動態的。如果尺寸不對，試試看用「比腳趾還小」這種描述，並多跑幾張圖。AI 的想像力也是樂趣的一部分！）
Description
Krea2 Giantess Concept LoRA (下面有中文介紹)
Krea2 base model already handles most giantess scenarios well without any LoRA. What it couldn't do is insertion of the mini-person — so I made this LoRA to fill that gap.
Trained on RAW to minimize overfitting and preserve Krea2's native diversity. Despite using anime-style素材, it can produce semi-photorealistic results with some luck.
Recommended weight: 1
Minor overfitting can be mitigated with precise prompts — e.g. specifying hair style/accessories to avoid matching the base look.
Known limitation: insertion only works in sitting or lying poses. Standing poses couldn't be trained due to lack of素材 — but that angle is rarely useful anyway, so it stays as-is.
Training on Krea2 is resource-heavy and costly to iterate, so don't expect frequent updates.
Credits:
- Distilled quantized models for inference — thanks to the authors
- Base model: DasiwaKrea2TurboRaw_mirroredskiesV1Raw.safetensors
- Test model: DasiwaKrea2TurboRaw_cutedisasterV2Turbo.safetensors (int8)
- Workflow adapted from Legendaer
- Prompt enhancement: ComfyUI-MieNodes
- And all other uncredited authors
============================================
Krea2 Giantess 概念 LoRA
Krea2 基底模型本身就對巨大娘題材有很好的理解，多數場景不用 LoRA 也能生成。唯獨「縮小人插入體內」這個動作始終出不來，所以做了這支 LoRA 補足這點。
使用 RAW 基底訓練，盡量保留 Krea2 原本的多樣性避免過擬合。雖然素材都是動漫風格，但勉強能出寫真感。
建議權重：1
微過擬合處可透過明確提示詞消除，例如指定髮型髮飾就能避免跟模型長一樣。
已知限制：縮小人插入的動作僅限坐姿或趴姿，站姿因缺乏訓練素材無法穩定產出——不過站姿插著的構圖實際上也較少用到，就先這樣了。
Krea2 訓練成本真的很高，短期內不太會更新。
特別感謝：
- 出圖可使用蒸餾量化版，感謝相關作者
- 訓練基底：DasiwaKrea2TurboRaw_mirroredskiesV1Raw.safetensors
- 測試用模型：DasiwaKrea2TurboRaw_cutedisasterV2Turbo.safetensors (int8)
- 測試工作流修改自 Legendaer
- 提示詞增強使用 ComfyUI-MieNodes
- 以及其他未具名的作者們