Boogu-Storyboard-Triptych-LoRA
Boogu Storyboard Triptych LoRA is an image-editing LoRA for Comfy-Org/Boogu-Image. Given one reference image and a structured Chinese instruction, it creates a vertical three-panel storyboard with a continuous action sequence while attempting to preserve the subject, clothing, props, environment, and visual style.
Boogu 三宫格叙事分镜 LoRA 是基于 Boogu-Image Edit 训练的图像编辑 LoRA。输入一张参考图与统一格式的中文提示词后，模型会生成一张自上而下排列的三宫格分镜，延续参考图中的人物、服装、道具、环境与画面风格，并表现连续动作。
Model file / 模型文件
boogu_storyboard_triptych_lora.safetensors- LoRA strength / 强度：
1.0 - Base model / 基础模型：
Comfy-Org/Boogu-Image - Recommended output / 推荐输出：
1024 × 2048 - Training checkpoint / 训练检查点：step
8000, epoch307 - Training framework / 训练框架：
ai-toolkit 0.10.20 - SHA-256：
ddab7bc679714d628da4c5eb0a09ca0fcbd2c3839c0eb3ceb8d52194860cdea8
Trigger phrase / 触发词
The full trigger phrase must be included at the beginning of every prompt:
每条提示词开头必须完整加入：
根据图一生成三宫格叙事连贯的镜头
Prompt standard / 提示词统一格式
根据图一生成三宫格叙事连贯的镜头，原图是[主体身份、固定外观、环境、起始动作]。后续镜头为[第一个连续动作]，随后[第二个动作或明确结果]。
Recommended writing rules / 推荐写法：
- Describe only visible subject identity, clothing, props, environment, and opening action in the
原图是...clause. - Use exactly two chronological and visually clear action beats after
后续镜头为.... - Keep the same subject, clothing, and key props throughout the instruction.
- Avoid unrelated style or quality keywords; the reference image provides the visual style.
- Keep the action causal and easy to divide into three frames.
Recommended parameters / 推荐参数：
| Parameter | Value |
|---|---|
| Width | 1024 |
| Height | 2048 |
| Steps | 6 |
| CFG | 1.0 |
| Sampler | Euler |
| Scheduler | Simple |
| Denoise | 1.0 |
| AuraFlow shift | 3.16 |
| LoRA strength | 1.0 |
LoRA comparison / LoRA 效果对照
The following 20 evaluations use the same reference image, prompt, random seed, base model, and sampling parameters for both outputs. The only variable is whether this LoRA is loaded.
以下 20 组测试中，无 LoRA 与有 LoRA 版本使用了完全相同的原图、提示词、随机种子、基础模型和采样参数，唯一变量是是否加载本 LoRA。
Description
Attention: This LoRA is made for Boogu Image Edit, not Krea2 Image.
Boogu-Storyboard-Triptych-LoRA
Boogu Storyboard Triptych LoRA is an image-editing LoRA for Comfy-Org/Boogu-Image. Given one reference image and a structured Chinese instruction, it creates a vertical three-panel storyboard with a continuous action sequence while attempting to preserve the subject, clothing, props, environment, and visual style.
Boogu 三宫格叙事分镜 LoRA 是基于 Boogu-Image Edit 训练的图像编辑 LoRA。输入一张参考图与统一格式的中文提示词后，模型会生成一张自上而下排列的三宫格分镜，延续参考图中的人物、服装、道具、环境与画面风格，并表现连续动作。
Model file / 模型文件
boogu_storyboard_triptych_lora.safetensors- LoRA strength / 强度：
1.0 - Base model / 基础模型：
Comfy-Org/Boogu-Image - Recommended output / 推荐输出：
1024 × 2048 - Training checkpoint / 训练检查点：step
8000, epoch307 - Training framework / 训练框架：
ai-toolkit 0.10.20 - SHA-256：
ddab7bc679714d628da4c5eb0a09ca0fcbd2c3839c0eb3ceb8d52194860cdea8
Trigger phrase / 触发词
The full trigger phrase must be included at the beginning of every prompt:
每条提示词开头必须完整加入：
根据图一生成三宫格叙事连贯的镜头
Prompt standard / 提示词统一格式
根据图一生成三宫格叙事连贯的镜头，原图是[主体身份、固定外观、环境、起始动作]。后续镜头为[第一个连续动作]，随后[第二个动作或明确结果]。
Recommended writing rules / 推荐写法：
- Describe only visible subject identity, clothing, props, environment, and opening action in the
原图是...clause. - Use exactly two chronological and visually clear action beats after
后续镜头为.... - Keep the same subject, clothing, and key props throughout the instruction.
- Avoid unrelated style or quality keywords; the reference image provides the visual style.
- Keep the action causal and easy to divide into three frames.
Recommended parameters / 推荐参数：
| Parameter | Value |
|---|---|
| Width | 1024 |
| Height | 2048 |
| Steps | 6 |
| CFG | 1.0 |
| Sampler | Euler |
| Scheduler | Simple |
| Denoise | 1.0 |
| AuraFlow shift | 3.16 |
| LoRA strength | 1.0 |
LoRA comparison / LoRA 效果对照
The following 20 evaluations use the same reference image, prompt, random seed, base model, and sampling parameters for both outputs. The only variable is whether this LoRA is loaded.
以下 20 组测试中，无 LoRA 与有 LoRA 版本使用了完全相同的原图、提示词、随机种子、基础模型和采样参数，唯一变量是是否加载本 LoRA。