Stable Scripts UI is a web-based user interface for Stable Scripts - a kind of python scripts can easily reproduce the same results created by other users to run GenAI models.
Comparing to other stable diffusion / GenAI UIs, SSUI has 5 major pros:
- Easy to use: Quickly create, run, and share stable scripts, which is self-contained and can automatically download the necessary pyt DDD7 hon modules, AI models and other dependent data.
- Reproducible: Scripts, its necessary modules, and the SSUI itself are versioned. Scripts can be run in the exactly same environment as the script author.
- Management: Manage your models and configurations in a centralized place.
- Strong Typed: All resources (include models) are strong types and can be prevented from being misused.
- Customizable: You can customize the scripts, types, models, panels, loaders, ui framework and more.
Currently, most of the GenAI tools are node-based, which is not friendly for large scale workflows. SSUI provides a project management system, which contains all the necessary information to run the script.
With SSUI, you can copy the project and shared with your friends. They can easily reproduce the results with all necessary resources.
SSUI provides a model management system, can help you easily download and install models from Civitai, HuggingFace and local files.
Stable Scripts provides good integration ability, which can allow you to use in different scenarios. Once a script is written, it can be called by:
- Other scripts
- In functional UI
- In Canvas - if the input and output of the function contains Images.
- In other extensions
We provides a VSCode plugin to help you write Stable Scripts. You can edit the code, and run the script inside VSCode.
Please refer to Stable Scripts document for details.
Currently supported models:
- SD1
- SDXL
- Flux
A Stable Script must contains 3 parts:
from ssui import workflow, Prompt, Image, Noise
from ssui_image.SD1 import SD1Model, SD1Clip, SD1Latent, SD1Lora, SD1Denoise, SD1LatentDecode, SD1IPAdapter
from ssui.config import SSUIConfig
from typing import List, Tupleconfig = SSUIConfig()3. Define a workflow function, with all the necessary type hints. And those type hints will be used to generate the UI.
@workflow
def txt2img(model: SD1Model, positive: Prompt, negative: Prompt) -> Image:
positive, negative = SD1Clip(config("Prompt To Condition"), model, positive, negative)
latent = SD1Latent(config("Create Empty Latent"))
latent = SD1Denoise(config("Denoise"), model, latent, positive, negative)
return SD1LatentDecode(config("Latent to Image"), latent)The config object will define each step in the workflow, you must pass it as the first argument of each API calls:
latent = SD1Latent(config("Create Empty Latent"))If there are configurable parameters, those controls will be generated in the details panel in the UI.
First, please check out the dependencies are installed on your device.
-
nodejs and yarn
- Please download from: https://nodejs.org/en/download
- npm install --global yarn
-
rustc and cargo
- Please check out https://www.rust-lang.org/tools/install
The following command will install yarn packages and a embedded python environment in '.venv' and necessary python packages for you.
yarnTo quickly start the development, you can run desktop project as the entry point:
yarn dev:desktopOr start additional hot reload servers for Functional UI development:
yarn dev:desktop
yarn dev:functional_uiOr you can start the server, executor, functional ui and desktop all manually, which can give you more flexibility, and clear messages:
yarn dev:server
yarn dev:executor
yarn dev:functional_ui
yarn dev:desktop- Application server: http://localhost:7422/
- API documentation server: http://localhost:7422/docs
- Functional UI dev server: http://localhost:7420/
If you are working on the functional ui, you can run it directly from the web browser:
yarn dev:server
yarn dev:executor
yarn dev:functional_uiExample URL:
http://localhost:7420/?path=<example_path>/basic/workflow-sd1.py
To test the image generation, you need the following models, including:
- SD1.5
- SDXL
- Flux
You can download them from: https://huggingface.co/datasets/sunxfancy/TestSDModels/tree/main
We use pre-commit to check the code before commit.
git config --local core.hooksPath .githooks/