[Feature Request]: Context-sensitive Styles list

[PaulCoddington]

Is there an existing issue for this?

• I have searched the existing issues and checked the recent builds/commits

What would your feature do ?

Problem:

1. Styles list ends up being used to remember trigger words for LoRA as well as preferred prompts for Models.

2. Styles list becomes unmanageably long and is unsearchable.

3. Having a large collection of LoRA, etc, requires some means of keeping track of Trigger Word lists, and/or non-trivial combinations of Trigger Words (in the case of LoRA with multiple characters).

Proposed feature is to implement basic support for context-sensitive Style lists.

Proposed workflow

Setup

1. Use the current Styles.csv as a global Styles list.

2. Allow per-model Styles.csv files to be located in the models and additional networks folders to be picked up automatically, eg: myModel.safetensors, myModel.styles.csv, myModel.vae.pt, myModel.png

Implementation

1. Content of the global Styles list always appears in the Style list for all models and additional networks.

2. Content of Styles associated with currently selected model and any LoRA (etc) added to prompt are appended to the Styles list.

User

1. Select model, as usual.

2. Select LoRA (etc) to add to prompt, as usual.

3. Select Styles that contain trigger words for LoRA (etc) from easily navigated shorter list, unencumbered by Styles that are only used with other models.

Additional information

It may be worth considering whether the Styles concept should include additional settings, such as Clip Skip preferred for current model, etc.