[Feature] Simple UI Mode

[iwr-redmond]

Congratulations on the progress being made in this repository. It won't be long before you're painting the town and the front porch!

A simple UI mode, which hides all but the most important options and allows users to explore at their own pace would make onboarding easier for new users - especially those who might balk at the comparative complexity of installing packages like SD.Next or SwarmUI (let alone ComfyUI!!).

Following the pattern of Fooocus and its forks, this would probably involve:

1. An initial "welcome" screen, offering to download some pre-determined models (based on presets) and allowing users to select simple or full UI mode
2. In simple mode, using generation presets to autoselect the best generation settings, like sampler and VAE, while providing users with the option to add Loras (see, e.g. Fooocus Lite for an excellent example of such beautiful simplicity) and probably set the number of steps ("Quality" or "Speed", for example)
3. In simple mode, having checkboxes that give easy access to extra functions - with optional Duke Nukem reference if desired!
4. In both modes, a resolution selector option (4:3, 16:9, etc.) that adapts to the native or upscaled model resolution (e.g. 4:3 = 512x512 for SD and 1024x1024 for SDXL)
5. A preset editor that allows users to share their defaults with others - you could adopt something like the SimpleSDXL or OG Fooocus format for this

[fszontagh]

I really like the idea! At the very least, I need to make some changes to the GUI layout. I always get stuck when trying to add something new to the GUI due to the limitations of wxWidgets' GUI builder and the additional constraints caused by cross-platform compatibility.

The concept you described reminds me of PrusaSlicer, where visible settings change based on the user's experience level.

However, if I want to implement a 'Simple / Intermediate / Expert' GUI mode, I may need to completely rewrite the underlying logic of the codebase.

And there are many bugs and limitations in the library itself that prevent the creation of a truly user-friendly GUI.

Currently, LoRAs can only be handled through prompts, meaning there’s no direct control over them in the GUI. There’s also no support for large embeddings. Another major issue is error handling—while the library logs errors, it doesn’t properly capture critical ones, such as 'Not enough VRAM' errors, meaning these cannot be shown to the user.

Fatal errors are also handled poorly, often just calling abort(), which is why the GUI runs the library in a separate process to prevent crashes from affecting the entire application.

On forward, i'd like to create some online service for these features, like downloadable prompt style, presets, predefined / random prompts.

[iwr-redmond]

I reckon that starting by making a list of what is feasible without burning too much daylight would be a good start, to help clarify what is possible at this stage.

[iwr-redmond]

The concept you described reminds me of PrusaSlicer, where visible settings change based on the user's experience level.

However, if I want to implement a 'Simple / Intermediate / Expert' GUI mode, I may need to completely rewrite the underlying logic of the codebase.

You could have an "Easy Mode" tab, which I reckon wouldn't be quite as disruptive for the existing code-base. Following the pattern of diffusers, you could then copy-paste code rather than creating complex abstractions to serve multiple purposes.