Stop At last layers of CLIP model/CLIP skip

[wmooring]

What does this option actually do? I've seen a lot a people referencing it and setting values other than the default in their generations, but there seems to be very little documentation on what it does and how it's supposed to work.

I assume it has to do with the language interpretation model? Is setting a higher value somewhat similar to setting a lower CFG value, where you'll get results that are less strictly adherent to your prompt but potentially more polished-looking?

Is there a reason so many people seem to be leaving it set to "2" now?

[aoirusann]

See wiki: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#ignore-last-layers-of-clip-model
Shortly, some models may work better with "2". (e.g. NAI)

[wmooring]

Ah thanks, I was searching for something actually called "Stop At last layers of CLIP model". It would be nice if it were named the same way in the wiki as in the settings interface.

That still doesn't really explain what the advantage of changing it would be. By "models", does it mean the .ckpt file or the sampler, or what?

[danzelus]

How make quick settings to this? Like sd_model_checkpoint or inpainting_mask_weight

[hd-x]

CLIP_stop_at_last_layers