Tips: English version is at the bottom
For online using, checkout the HuggingFace Space
Fooocus V2动态提示功能的Webui移植，安装之后首次运行会自动从 https://huggingface.co/lllyasviel/misc/resolve/main/fooocus_expansion.bin 下载GPT-2模型，并保存到插件目录下的 models/expansion/pytorch_model.bin
如果自动下载失败，可以自己手动下载。
安装成功只要勾选启动就会生效。
启用前：
启用后：
关于Fooocus V2原版功能介绍： lllyasviel/Fooocus#117 (comment)
新增 SuperPrompt V1 https://huggingface.co/roborovski/superprompt-v1
由于huggingface_hub/transformers的下载功能不稳定，需要手动clone到插件目录下的 models/superprompt-v1
可以通过启动参数 --pe-model-path 来指定模型目录
两种模型的区别：
- Fooocus V2: 只会在原提示词的结尾添加一些画质相关的提示词，不会对原提示词进行修改，因此会忠于原提示词的画面风格
- SuperPrompt V1: 会对原提示词进行大幅重写，可能会得到非常不同的画面效果
新增 DanTagGen-beta https://huggingface.co/KBlueLeaf/DanTagGen-beta
由于huggingface_hub/transformers的下载功能不稳定，需要手动clone到插件目录下的 models/DanTagGen-beta
可以通过启动参数 --pe-model-path 来指定模型目录
本模型适用于 Kohaku-XL 等使用 danbooru 标签的模型
- 修复了 SuperPrompt V1 输出不受种子值影响的问题
- 实现了 DanTagGen-beta 的渐进生成机制，扩展的提示词会更丰富
- 在
Advanced Option中提供了自定义 SuperPrompt V1 使用的提示词的配置
- 将 DanTagGen 支持改为 HF collections 里的任意版本
WebUI port of Fooocus V2 dynamic prompt expansion. Upon installation and first run, it automatically downloads the GPT-2 model from https://huggingface.co/lllyasviel/misc/resolve/main/fooocus_expansion.bin and saves it to models/expansion/pytorch_model.bin in the plugin directory.
If the automatic download fails, you can download it manually.
Once installed, simply checking the box to activate it will make it take effect.
Before enabling:
After enabling:
For more information on the original Fooocus V2 feature, visit: lllyasviel/Fooocus#117 (comment)
Added SuperPrompt V1 https://huggingface.co/roborovski/superprompt-v1
Due to the unstable download function of huggingface_hub/transformers, it needs to be manually cloned to models/superprompt-v1 in the plugin directory.
The model directory can be specified via the --pe-model-path startup parameter.
Differences between the two models:
- Fooocus V2: Only adds some image quality-related prompt words at the end of the original prompt, without modifying the original prompt, thus maintaining the original artistic style.
- SuperPrompt V1: Significantly rewrites the original prompt, potentially resulting in very different visual effects.
Added DanTagGen-beta https://huggingface.co/KBlueLeaf/DanTagGen-beta
Due to the unstable download function of huggingface_hub/transformers, it needs to be manually cloned to models/DanTagGen-beta in the plugin directory.
The model directory can be specified via the --pe-model-path startup parameter.
This model is suitable for models that use danbooru tags, such as Kohaku-XL.
- Fixed an issue where SuperPrompt V1 output was not affected by seed values.
- Implemented a progressive generation mechanism for DanTagGen-beta, resulting in richer expanded prompt words.
- Added a custom configuration for prompt words used by SuperPrompt V1 in the
Advanced Option.
- Changed DanTagGen support to any version in HF collections
- Added sampling parameters in Advanced Option, so that you can control the output randomness.