[qwen-image-edit-2509]AniSockChanger

基于qwen image edit 2509 BF16训练的换袜子功能模型。

基础公式: 换成 [装饰/颜色] [袜子类型]

叠穿公式: 换成 [底层] 套 [中层] 套 [表层]

▶️ 单层示例:

• 换成白色过膝丝袜

• 换成荷叶边白色短袜

• 换成黑色连裤丝袜

• 换成吊带白色过膝丝袜

▶️ 叠穿示例:

• 换成白色连裤丝袜套黑色过膝丝袜

• 换成白色连裤丝袜套荷叶边白色短袜

• 换成黑色连裤丝袜套吊带白色过膝丝袜套荷叶边白色短袜

▶️ 特殊状态示例:

• 换成半脱白色连裤丝袜

• 换成裆部破损黑色连裤丝袜

• 换成内裤外穿白色连裤丝袜

推荐设置

• LoRA权重: 1.0

• 请严格遵循 "换成..." 的句式，以获得最可控的效果

• 推荐使用轻量4步加速模型生成Qwen-Image-Edit-2509-Lightning-4steps-V1.0-bf16_dim1.safetensors2vXpSwA7/iroiro-lora at main

下个版本将会继续扩充训练集，并支持袜子状态-裸腿状态的编辑(貌似模型原生即可支持？)

第一次发布模型，如果有不足之处欢迎指正~希望多多收集经验和意见

======================================================
A functional LoRA for changing socks, trained on Qwen Image Edit 2509 (BF16).

V1 supports editing from a bare-legged state to a state with socks on. It was trained with 176 image pairs at 1024 resolution using Chinese tags.

Basic Formula: 换成 [decoration/color] [sock type]

Layering Formula: 换成 [Bottom Layer] 套 [Middle Layer] 套 [Top Layer]

▶️ Single-Layer Examples:

换成白色过膝丝袜 (change to white over-the-knee socks)

换成荷叶边白色短袜 (change to white short socks with frills)

换成黑色连裤丝袜 (change to black pantyhose)

换成吊带白色过膝丝袜 (change to white garter over-the-knee socks)

▶️ Layering Examples:

换成白色连裤丝袜套黑色过膝丝袜 (change to white pantyhose with black over-the-knee socks on top)

换成白色连裤丝袜套荷叶边白色短袜 (change to white pantyhose with frilly white short socks on top)

换成黑色连裤丝袜套吊带白色过膝丝袜套荷叶边白色短袜 (change to black pantyhose + white garter OTK socks + frilly white short socks)

▶️ Special State Examples:

换成半脱白色连裤丝袜 (change to partially removed white pantyhose)

换成裆部破损黑色连裤丝袜 (change to crotch-torn black pantyhose)

换成内裤外穿白色连裤丝袜 (change to panties over white pantyhose)

Recommended Settings

LoRA Weight: 1.0

Please strictly follow the "换成..." prompt structure for the most predictable results.

It is recommended to use the lightweight 4-step acceleration model: Qwen-Image-Edit-2509-Lightning-4steps-V1.0-bf16_dim1.safetensors

The next version will expand the training set and support editing from a socked state to a bare-legged state (it seems the base model might already support this?).

This is my first time releasing a model. Any feedback or suggestions for improvement are welcome! I hope to gather more experience and advice.