# 模型说明 (Model Description)

本 LoRA 模型基于 Wan2.2 视频生成模型训练而成，使用了 20 个视频作为训练数据，总共训练 2000 步。

This LoRA model is trained on the Wan2.2 video generation model, using 20 videos as the dataset with a total of 2000 training steps.

# 底模区分 (Base Model Separation)

针对 Wan2.2 high noise 和 Wan2.2 low noise 两个底模，分别训练了独立的 LoRA 模型。

Separate LoRA models were trained for the Wan2.2 high noise and Wan2.2 low noise base models.

# 使用建议 (Usage Recommendations)

- 不建议在推理时使用 Lighting 加速，会严重影响视频生成效果。

Lighting acceleration is not recommended during inference, as it significantly degrades video quality.

- 除了触发词外，可以在提示词中加入额外的限制条件，例如花瓣颜色、数量、大小等，以获得更可控的生成效果。

Beyond the trigger words, you may add additional constraints in the prompt (e.g., petal color, amount, size) to achieve more controllable results.

# 适用场景 (Applicable Scenarios)

- High noise LoRA：更适合动态效果和大幅度变化（如人物瓦解成花瓣、强烈动作）。

High noise LoRA: better suited for dynamic effects and drastic changes (e.g., character dissolving into petals, strong motions).

- Low noise LoRA：更适合保持细节和局部修改（如人脸结构、服饰纹理）。

Low noise LoRA: better suited for detail preservation and localized edits (e.g., face structure, clothing textures).

# 提示词扩展 (Prompt Enhancements)

- 颜色控制：如 "red petals, no white petals"

Color control: e.g., "red petals, no white petals"

- 数量与大小：如 "thousands of tiny petals, evenly scattered"

Quantity & size: e.g., "thousands of tiny petals, evenly scattered"

- 动态过程：如 "gradually dissolve into petals, petals drifting with wind"

Dynamic process: e.g., "gradually dissolve into petals, petals drifting with wind"

- 环境保持：如 "background untouched, petals only from subject"

Environment preservation: e.g., "background untouched, petals only from subject"