Skip to content

Extensions

Jump to bottom
w4ffl35 edited this page May 14, 2023 · 15 revisions

Coming soon

Extensions are a work in progress. Keep an eye on this page for updates on how to create extensions.

Although a previous version of AI Runner contained extensions, only a LoRA extension has been released. The decision was made to move LoRA into the AI Runner code-base and remove extensions until the extension feature is more robust and mature.

Extension class

Extensions must contain a main.py file which extends the BaseExtension class.

Example

from airunner.extensions import BaseExtension


class Extension(BaseExtension):
    ...

Injection methods

You can modify various GUI elements by adding specific injection methods to your Extension class.

Generator tab

Generator tab refers to widget which contains all of the elements found in each generator section

Use the generator_tab_injection function to manipulate the Generator tab

Example

def generator_tab_injection(self, tab: QWidget, tab_name: str = None):
   # modifications here
   ...

See the LoRA extension for an extensive example

generator_injection

Menubar

The menubar sits at the very top of the main window.

Use the menubar_injection function to manipulate the menubar.

Example

class Extension(BaseExtension):
    def menubar_injection(self, menubar: QMenu):
        menubar.addMenu("Menu Example")

menubar_injection

Toolbar

The toolbar sits just under the menubar and contains various buttons such as Undo and Redo.

Use the toolbar_injection function to manipulate the toolbar.

Example

def toolbar_injection(self, toolbar: QFrame):
    self.example_button = QPushButton("Button Example")
    toolbar.layout().addWidget(self.example_button)

toolbar_injection

Generate data injection

When generate is called, AI Runner passes form data from the generate tab to a class which is responsible for handling Stable Diffusion generation. The data which is passed to this class can be manipulated with the generate_data_injection function.

Example

def generate_data_injection(self, data):
    data["options"]["lora"] = []
    for lora in self.available_loras(data["action"]):
        if lora["enabled"]:
            print(lora)
            data["options"]["lora"].append((lora["name"], lora["scale"]))
    return data

Image generation manipulation

the call_pipe method can be used to gain access to the stable diffusion model.

def call_pipe(self, options: dict, model_base_path: str, pipe: object, **kwargs):
   ...
   return pipe

See the LoRA extension for an extensive example

Extension submissions

In order to submit an extension for review, send a email to [email protected] with a link to your extension on github.

  • Extensions must be released on Github under the AGPL-3.0 license.
  • A setup.py file must exist in the root directory
  • You must create a release

Example setup.py

setup(
    name='airunner-lora',
    version="1.0.0",
    author="Capsize LLC",
    description="LoRA extension for AI Runner",
    license="AGPL-3.0",
    author_email="[email protected]",
    url="https://github.com/Capsize-Games/airunner-lora",
)

See the airunner-lora extension for further examples.

Clone this wiki locally