sd-random-prompt-generator

日本語のReadmeはこちら。

sd-random-prompt-generator is the image generator with random prompts via the API of Stable Diffusion web UI. You can generate random characters, random emotions, random backgrounds and random poses❗

This generator mainly targets SD 1.5. But, maybe it can be used for SDXL.

You need to edit TypeScript codes. If you have no skill of any programming language, this tool may be difficult to use...

✨ Features

• Build the random prompts from the settings.
• Generate image with the random prompts via the API of Stable Diffusion web UI.
  • No need to open the web UI on your web browser.
• Export random prompts to text.
  • It supports Dynamic Prompts style. You can copy-paste it to prompts area of web UI.

🔌 Requirements

• Node.js
• Stable Diffusion web UI
  • Extensions: sd-webui-cutoff
  • Checkpoints, VAE, Textual Inversions, Loras, etc...

🍎 Getting Started

Enable API on the web UI

Add --api options to COMMANDLINE_ARGS in webui-user.bat like below.

set COMMANDLINE_ARGS=--listen --xformers --api

Setup Cutoff extensions on the web UI

See the doc .

I strongly recommend to use this extension. But, if you don't need it, comment out the following codes.

• src/generator.mts
  • generateEachImage()

    alwayson_scripts: {
      cutoff: {
        args: [
          true,
          "red,pink,orange,yellow,green,blue,purple,black,white,brown,grey,blonde",
          1.5,
          false,
          false,
          "",
          "Lerp",
          true,
        ],
      },
    },

Install npm packages

npm run install

Edit setting.mts

You need to edit the following file.

• src/setting.mts

Change machineSetting as you like.

export const machineSetting = {
  ip: `192.168.10.3`,
  port: 7860,
} as const satisfies MachineSetting;

Change settings as you like.

export const settings = [
  {
    key: `all`,
    fixedPrompt: `1girl, solo, masterpiece, best quality,\n`,
    optionsBodyJson: {
      sd_model_checkpoint: `vividorangemix_v10.safetensors [ff4725f91c]`,
      sd_vae: `blessed2.vae.safetensors`,
    },
    txt2imgBodyJson: {
      negative_prompt: `verybadimagenegative_v1.3, (cameltoe:1.5)`,
      sampler_name: `Restart`,
      width: 512,
      height: 768,
      cfg_scale: 5,
      denoising_strength: 0.4,
      enable_hr: true,
      hr_scale: 2.5,
      hr_upscaler: "4x-AnimeSharp",
      hr_second_pass_steps: 30,
    },
    characters: [
      {
        key: `amaburi-isuzu-little-jelly`,
        probability: 1,
        outfits: [
          {
            key: `bikini`,
            probability: 1,
            backgrounds: [
              {
                type: `from-horizontal`,
                key: `ocean`,
                poses: [{ key: `portrait` }, { key: `arms-up` }],
              },
              {
                type: `from-above`,
                key: `bed-sheet`,
                poses: [{ key: `lying-on-bed` }, { key: `lying-on-stomach` }],
              },
            ],
          },
        ],
      },
    ],
  },
] as const satisfies Setting[];

Top level settings.

Key	Meaning
key	Just used for filename. Don't affect generation.
fixedPrompt	Fixed prompt. It attached to top of your prompts.
optionsBodyJson.sd_model_checkpoint	Checkpoint (*1).
optionsBodyJson.sd_vae	VAE (*1).
txt2imgBodyJson.negative_prompt	Negative prompts (*1).
txt2imgBodyJson.sampler_name	Sampler (*1).
txt2imgBodyJson.width	Image width (*1).
txt2imgBodyJson.height	Image height (*1).
txt2imgBodyJson.cfg_scale	CFG Scale (*1).
txt2imgBodyJson.denoising_strength	Upscaler denoising strength (*1).
txt2imgBodyJson.enable_hr	Upscaler Enable/Disable.
txt2imgBodyJson.hr_sclae	Upscaler scale (*1).
txt2imgBodyJson.hr_upscaler	Upscaler name (*1).
txt2imgBodyJson.hr_second_pass_steps	Upscaler steps (*1).
characters	Character settings. See below.

Character settings. You can define multiple characters that is picked by probability when generating.

Key	Meaning
key	Character key (*2).
probability	Probability weight. Optional. Default is 1 .
outfits	Outfit settings. See below.

Outfit settings. You can define multiple outfits that is picked by probability when generating.

Key	Meaning
key	Outfit key (*2).
probability	Probability weight. Optional. Default is 1 .
backgrounds	Background settings. See below.

Background settings. You can define multiple outfits that is picked by probability when generating.

Key	Meaning
type	Camera type. It determines which poses can be used (*2).
key	Background key (*2).
probability	Probability weight. Optional. Default is 1 .
poses	Pose settings. See below.

Pose settings. You can define multiple poses that is picked by probability when generating.

Key	Meaning
key	Pose key (*2).
probability	Probability weight. Optional. Default is 1 .

(*1): Extend setting defines

Some settings can only be set to defined values.

If you want to extend (add) defines, edit the following file.

Setting	Path	Variable or Type define
optionsBodyJson.sd_model_checkpoint	src/setting-define.mts	type Checkpoint
optionsBodyJson.sd_vae	src/setting-define.mts	type VAE
optionsBodyJson.sampler_name	src/setting-define.mts	type Sampler
optionsBodyJson.width	src/setting-define.mts	type Setting
optionsBodyJson.height	src/setting-define.mts	type Setting
optionsBodyJson.denoising_strength	src/setting-define.mts	type Setting
optionsBodyJson.hr_scale	src/setting-define.mts	type Setting
optionsBodyJson.hr_upscaler	src/setting-define.mts	type Upscaler
optionsBodyJson.hr_second_pass_steps	src/setting-define.mts	type Setting

(*2): Character, outfit, background, pose keys (types)

Each key is used for determing which defined data set is used. If you want to know the key list, see the following files.

Key	Path	Variable or Type define
Character key	src/characters/resolver.mts	variable characterTable
Outfit key	src/outfits/resolver.mts	variable outfitTable
Background type	src/backgrounds/resolver.mts	type BackgroundType
Background key	src/backgrounds/resolver.mts	variable backgroundTable
Pose key	src/poses/resolver.mts	variable poseTable

Basically, the character settings depend on Lora. You need to install these Loras. What Loras do you need is described the following files.

• src/tag-defines/lora.mts

Generate and Exports

npm run exe

The tool builds prompts from your setting. Generate images and asynchronously exports prompts to outputs directory.

The tool periodically checks progress. You will see something like the following in your terminal.

  10 s: [=========>          ]  49%
  15 s: [===========>        ]  56%
  17 s: [============>       ]  62%

🔧 Define characters, outfits, backgrounds and poses as you like

🚧 WIP 🚧

📖 Tech stack

• TypeScript
• tsx