Batch creation with step and guidance scale intervals

[ChrisBii]

Running Latest Version

• I am running the latest version

What do you want Mochi Diffusion to do?

It would be nice to have the ability to do batches with interval step count and guidance scale.

Example:
Generate 20 images with step count between 10-30
Generate 13 images with guidance scale between 7,0 - 12,0

Why do you think this should be added?

Setting step count and guidance scale each time to find the best combo manually is incredibly tedious.

[jrittvo]

I would use this kind of feature often. Perhaps this could be implemented in an "Automate" tab in the "Settings" panel. In essence, the "Automate" tab would be the home for batch run settings. The user who has no need for that kind of thing can just ignore it.

The "Automate" tab would need an "enable/disable" toggle so that when enabled, the "NUMBER OF IMAGES" slider in the main interface was disabled, in this initial use case. Other use cases might need other overrides.

Other ideas for an "Automate" tab:

• If the number of available schedulers were increased, "Automate" could cycle through them.
• Perhaps at some point, the VAE's baked into the models could be optionally replaced by alternate VAE's from a library of VAE's that the user populates. "Automate" could then cycle through VAE's as well.

I imagine adding in loops to increment numerical settings or to cycle through a list of arguments would be trivial. These kinds of things let the app accomplish things with just a few clicks that save the user a great deal of time fiddling with inputs to evaluate variations while fine tuning output.

If this were implemented, it would also be a good idea to add a "Settings" icon to the top tool bar for faster access.

[atemiz]

I've been thinking the same thing for the past few days. Sometimes generating the same image with more or less steps can give better results.
On the interface side, images generated with the same seed and prompt but different step count can be grouped. It would be nice if we could then see the differences in the desired step range (eg 10-15 & 17-20) with a slider in the preview window. This process can be more efficient if the result for each step during the generating process can be exported. However, if this is not possible with the current system, a queuing mechanism or at least a simple automation (which will increase the number of steps and start the generate process again) can work.