Ace-Step 1.5 & Kobold Lite QOL

[OldMan0000-sys]

First thank you for all the work you have done on this app, amazing stuff!

There are a couple of quality of life items that I want to bring up, maybe there is already a way to achieve these, I'm just not the technical sort.

1. For Ace-Step 1.5 when hitting generate without hitting plan first, non of the empty boxes will auto-fill as far as I can see in the output. Maybe this is by design, but can we have the option to auto-fill every box that's empty even if we don't hit plan first? This would also act as randomize everything not filled option for every generation.
2. For Ace-Step 1.5 if I have a entered a value in the boxes that value shouldn't change if I hit plan, some of the values get changed. Also some are just stuck between a few options, for example it always inputs "G Major" or "G Minor" in the scales.
3. I like the low vram option available for the audio generation. Can we have that option too for text and picture generation? Lets say I want to generate text, empty vram, generate a picture, empty vram, and then continue to generate text. This would let us use bigger models instead of trying to fit everything together with limited vram.
4. Is there a way to use 2 image generation models? Lets say I want to use z-image to generate & klein 9b to edit. vram management to enable this and also a way to switch form generate to edit in the same interface would be great. This would instantly get me off of Comfy for image generation at least.

[LostRuins]

can we have the option to auto-fill every box that's empty even if we don't hit plan first?

this is hard because the planner model and the generation model are different, it's possible to load one without the other. The things like BPM, duration, key, all come from the planner.

For Ace-Step 1.5 if I have a entered a value in the boxes that value shouldn't change if I hit plan, some of the values get changed. Also some are just stuck between a few options, for example it always inputs "G Major" or "G Minor" in the scales.

This happens as everything changes when lyrics are rewritten (e.g. longer lyrics need longer duration), in general, it's helpful to manually re-adjust some parameters after generating.

I like the low vram option available for the audio generation. Can we have that option too for text and picture generation?

Yes a mode for that will be added next version.

Is there a way to use 2 image generation models?

It's not automatic, however you can create 2 kcpps config files, and then use the admin endpoint to swap between them at runtime.
https://github.com/LostRuins/koboldcpp/wiki#what-is-admin-mode-can-i-switch-models-at-runtime