koishijs/novelai-bot

Feature: 关于重绘强度

ShanShuWu opened this issue · 1 comments

Scope

Stable Diffusion WebUI (AUTOMATIC1111)

Describe the problem related to the feature request

这个功能我输入-e提示我不能用增强功能,返回为 增强功能仅适用于被生成的图片。普通的 img2img 请直接使用「约稿」而不是「增强」。我想知道这个功能的具体使用方法,以图绘图很需要这个参数来让成品和原图尽量相似

Describe the solution you'd like

重绘强度,使出图和原图提高相似度(可能已经实现,但我实在不会用)

Describe alternatives you've considered

No response

Additional context

No response

这个问题不止一个,我想逐一回答。

Re: 重绘强度

关于 "重绘强度",我想你指的是初始图像和结果图像之间的差异,这在稳定扩散模型中被称为去噪强度

如果你输入help novelai,你会看到这样的输出。

novelai <prompts>
从Novel AI生成图像
别名:nai。
输入 "novelai "的英文提示或标签,例如,森林中的女孩,金发,红眼,白裙子,等等。
你也可以使用逗号分隔的标签,如Danbooru上的标签。
星星吧:https://github.com/koishijs/novelai-bot
可用的选项。
    -r, --resolution <resolution> 设置图像分辨率
    -o, --输出
    -O, --override 禁用默认提示语
    -s, --sampler <sampler> 设置采样器
    -x, --seed <seed> 设置随机种子
    -t, --steps <step> 设置迭代步数
    -c, --scale <scale> 设置CFG规模
+   -N, --strength <strength> 设置去噪强度
    -u, --undesired <undesired> 负面提示
    -i, --iterations <iterations> 设置批量计数。

所以你学会了如何控制它:只要在0~1中加入-N-strength的浮点数。

关于:增强模式

在这个错误信息上有一点过时了。增强模式只能用于生成的图像。如果你使用正常的img2img,请使用 "novelai",不要使用增强选项。
历史上我们只支持NovelAI的小说生成AI服务网站,所以这里的信息没有明确提到。事实上,只有当你使用NovelAI作为后台时,增强模式才可用。

关于:目前使用img2img的方式

这是非常简单和直接的。你向novelai <your-prompt>发送一张图片,然后你发送的图片将被用作初始图片送入稳定的扩散模型。
对于一些不支持文本-图像-混合信息的过时的聊天平台,你也可以先发送一个图像,然后用novelai命令quote(或reply)这个信息。

通过www.DeepL.com/Translator(免费版)翻译


The original text

There are more than one question in this issue, I would like to reply one by one.

Re: redraw strength

As Redraw strength, I am assuming you're referring the differences between the initial image and the result one, which is called denoising strength in Stable Diffusion models.

If you type help novelai, you would see the output like this:

novelai <prompts>
Generate Images from Novel AI
Aliases: nai.
Enter "novelai" with English prompt or tags, e.g. a girl in the forest, blonde hair, red eyes, white dress, etc.
You can also use comma separated tags like those on Danbooru.
Star it: https://github.com/koishijs/novelai-bot
Available options:
    -r, --resolution <resolution>  Set Image Resolution
    -o, --output
    -O, --override  Disable Default Prompts
    -s, --sampler <sampler>  Set Sampler
    -x, --seed <seed>  Set Random Seed
    -t, --steps <step>  Set Iteration Steps
    -c, --scale <scale>  Set CFG Scale
+   -N, --strength <strength>  Set Denoising Strength
    -u, --undesired <undesired>  Negative Prompt
    -i, --iterations <iterations>  Set Batch Count.

So you learnt how to controler it: just add -N or --strength with a float number in 0~1.

Re: enhance mode

There is a bit obsolete on this error message: The Enhance mode can only be used for images generated. Use "novelai" without enhance option if you are using normal img2img.
Historically we only support NovelAI the novel generation AI service website, so the message here didn't explicitly mention. In fact, the enhance mode is only available if you are using NovelAI as the backend.

Re: the current way to use img2img

It is very simple and straightforward. You send novelai <your-prompt> with an image, then the image you sent would be used as the initial image fed to stable diffusion model.
For some obsolete chat platform that don't support text-image-mixin messages, you could also send an image first, and then quote (or reply) the message with novelai command.