培訓的素材只有陰莖特寫，沒有訓練任何臉部，並且只有正面勃起陰莖的視角，素材只有大陰莖沒有小陰莖。 沒有訓練任何的性愛動作，皆是由minimax h3內建的功能生成。
我訓練正面勃起陰莖的視角已經失敗了多次，由於只靠一個lora無法生成陰莖的細節，所以我分為lora A和B版本，A版本訓練步驟步驟較低，B版本訓練步驟較高，建議A與B版本同時使用。
如果首幀或尾幀沒有良好的龜頭圖片，依然無法生成出完美的龜頭。
此lora只在”I2V”模式訓練。
LORA使用H100 SXM訓練，解析度為1536，具有詳細的陰莖紋理。
工作流只使用comfyui中官方模板的預設工作流，然後添加” minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16 lora”，並沒有使用其他任何複雜的工作流。
在"Training data"中有工作流.
Lora權重:0.75 to 0.9.
The training material consists exclusively of close-up shots of penises. No faces were included in the training, and the training data only covers a frontal view of an erect penis. The dataset contains only large penises; no small penises were included. No sexual activity or sexual actions were trained; all such movements are generated using the built-in capabilities of MiniMax H3.
I have failed multiple times when training the frontal view of an erect penis. Since a single LoRA cannot reproduce sufficient anatomical detail, I divided it into LoRA A and LoRA B versions. Version A uses fewer training steps, while Version B uses more training steps. It is recommended to use both A and B versions together.
If the first or last frame does not contain a good-quality image of the glans, it will still be difficult to generate a perfect glans.
This LoRA was trained exclusively in I2V mode.
The LoRA was trained on an H100 SXM at a resolution of 1536, providing detailed penile texture.
The workflow uses only the default workflow from the official ComfyUI templates, with minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16 lora added. No other complex workflows were used.
The workflow is included in the "Training data" section.
LoRA weight: 0.75–0.9
觸發詞(Trigger words):
A frontal view of an erect penis with the glans clearly visible. a man's erect penis from the front. prominent penile veins. His testicles were visible.
The man beneath her has his buttocks facing the camera with his legs spread.
Description
培訓的素材只有陰莖特寫，沒有訓練任何臉部，並且只有正面勃起陰莖的視角，素材只有大陰莖沒有小陰莖。 沒有訓練任何的性愛動作，皆是由minimax h3內建的功能生成。
我訓練正面勃起陰莖的視角已經失敗了多次，由於只靠一個lora無法生成陰莖的細節，所以我分為lora A和B版本，A版本訓練步驟步驟較低，B版本訓練步驟較高，建議A與B版本同時使用。
如果首幀或尾幀沒有良好的龜頭圖片，依然無法生成出完美的龜頭。
此lora只在”I2V”模式訓練。
LORA使用H100 SXM訓練，解析度為1536，具有詳細的陰莖紋理。
工作流只使用comfyui中官方模板的預設工作流，然後添加” minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16 lora”，並沒有使用其他任何複雜的工作流。
在"Training data"中有工作流.
Lora權重:0.75 to 0.9.
The training material consists exclusively of close-up shots of penises. No faces were included in the training, and the training data only covers a frontal view of an erect penis. The dataset contains only large penises; no small penises were included. No sexual activity or sexual actions were trained; all such movements are generated using the built-in capabilities of MiniMax H3.
I have failed multiple times when training the frontal view of an erect penis. Since a single LoRA cannot reproduce sufficient anatomical detail, I divided it into LoRA A and LoRA B versions. Version A uses fewer training steps, while Version B uses more training steps. It is recommended to use both A and B versions together.
If the first or last frame does not contain a good-quality image of the glans, it will still be difficult to generate a perfect glans.
This LoRA was trained exclusively in I2V mode.
The LoRA was trained on an H100 SXM at a resolution of 1536, providing detailed penile texture.
The workflow uses only the default workflow from the official ComfyUI templates, with minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16 lora added. No other complex workflows were used.
The workflow is included in the "Training data" section.
LoRA weight: 0.75–0.9
觸發詞(Trigger words):
A frontal view of an erect penis with the glans clearly visible. a man's erect penis from the front. prominent penile veins. His testicles were visible.
The man beneath her has his buttocks facing the camera with his legs spread.