Never done it myself, but this came up in a Google search from Reddit user VegaKH; If you are using Auto1111 you can easily convert models to SafeTensors without installing another piece of software.

1. Go to checkpoint merger tab.

2. Choose the model you want to convert in "Primary Model (A)"

3. Set the exact same model in "Secondary Model (B)"

4. Choose a custom name for the new model.

5. Set multiplier to 0. Leave "weighted sum" checked.

6. Choose safe tensors

7. Optionally choose FP16 to save space on the hard drive (may slightly degrade image quality, I see no difference.)

8. Hit Run.

@natewilliams does this still work? i can only choose checkpoints in a1111's checkpoint merger tab

[bad_pictures:0.8] IS working in auto1111 as stated on official wiki in github.
However its not a weight but sampling time to activate token.
for step count - 40 and [bad_pictures:0.8] lora will start to work at step (40*0.8=)32
Its not a bad idea to put [bad_pictures:0.2] so effect will refine details but not affect initial structural noise at first 5-10 steps

@freezerain Right, I should have been clearer about that. Prompt scheduling is a bit different than weighting, and (I think) it's possible to use both together: ([bad_pictures:0.5]:0.8) would give you bad_pictures, at a weight of 0.8, for the last 20 out of 40 steps. Weighting and prompt editing could have similar effects, but they're slightly different. Vivipeg stated specifically that they meant it as weighting, in which case the bracketed version would be incorrect, as I commented. Otherwise, though, that's a good point that the syntax itself isn't incorrect. It just isn't what Vivipeg intended.

@Vivipeg Ok, thanks