Open brknsoul opened 1 month ago
This is one of the main reasons I dont use SD.next as often, on auto1111 with civitai helper extension (it downloads the metadata of the lora) you can click a lora and it adds all the trigger words that are automatically put under the activation text on that screenshot.
I hope they add this feature because I love SD.next but is a lot of work to copy paste the trigger words of loras atm.
this is a very valid request, but what you're describing is already mostly in sdnext - it can download metadata from civitai and it places triggers/tags as you hover on lora in networks panel. and clicking on each adds/removes them from the prompt.
adding all triggers at once is not how most loras work as different triggers can be used to trigger different lora behavior.
Well on SD.next I notice that the trigger words that appear when you hover the lora are not even complete like in this screenshot (hovering on the top right lora Cloud).
I usually have to go to the Lora info and check the model metadata tab to read the comment of the autor to see if they put the trigger words in there, which some of they dont even do and I have to go to the model page and manually copy and paste.
What I have to do is copy the trigger words and add they on model description like on this screenshot, from there I can just open the model info and copy paste what I want, but it makes me wonder if this description tab can be used as the activation text like auto1111 so if I click a lora it just add whatever is in there to prompt.
Oh and this is how the activation text looks on auto when I use civitai helper to scan the metadata.
So I just click a button to scan and then refresh the networks and after that if I click a lora it will add all that to prompt without extra steps of copying and pasting.
adding all triggers at once
That's not what we're asking. Most loras have a trigger word that's always used. We'd like this to be added to the prompt when adding the lora.
We don't need all trigger words added.
I agree. adding ALL trigger words is bad in most of the cases, see what I wrote - they can (and do) refer to different parts of lora that are actually in conflict. you need to choose trigger words you want to use, not blindly add them all.
And if you state thar sdnext suggested trigger words are not complete, provide actual example (and with a link to lora).
Understandable to be honest this is just a personal preference since I really dont use the hover on lora trigger words because you need to click one by one and some loras have mora than 10 trigger words and when I try to click further down they disappear because of how many they are and you already hovering on another lora.
I will keep using the workaround of putting the trigger words on the description tab inside the info of the model.
i suggest to actual test if lora really needs 10+ trigger words or is 1-2 enough - in 99% its later. if lora really needs 10+ trigger words for a single style, that's really bad way to train a lora.
i want to use civitai provided trigger word not embedded metadata. a111 uses civitai provided trigger word but sdnext uses embedded metadata. is a1111's implementation incorrect? am i missing some setting to set civitai's trigger word as preferred over embedded metadata.
i want to use civitai provided trigger word not embedded metadata. a111 uses civitai provided trigger word but sdnext uses embedded metadata. is a1111's implementation incorrect? am i missing some setting to set civitai's trigger word as preferred over embedded metadata.
sdnext uses BOTH.
Feature description
It would be nice if the Lora info panel could be re-done with some of the features found in Forge/A1111, specifically;
Here is the panel from Forge for reference.
Version Platform Description
No response