Merges a few sliders and requires no trigger. A unique style. Examples at 0.7-1 strength. Create your own at sliders.ntcai.xyz
More info:
https://sliders.ntcai.xyz/sliders/app/merges/1cb89b00-b7d6-4d5e-8091-1705b6fa413f
A merge of captivating, Stunningly Beautiful, attention-grabbing, entrancing, attention-stealing then trained to remove the need for triggers.
Support my work on patreon and get access to over 1400 unseen loras and 11000 lora merges (more daily).
Patreons get: Create your own private sliders with text
Patreons get: Merge your own sliders or any of the public ones
Patreons get: A repository of over 1000 sliders
Patreons get: A repository of over 10000 merges
She stole the moon?
Description
FAQ
Comments (12)
The patreon is all well and good, but what does this slider actually do?
https://sliders.ntcai.xyz/sliders/app/merges/1cb89b00-b7d6-4d5e-8091-1705b6fa413f
Its this one but trained with an experimental technique that removes the need for triggers
@ntc Thank you. It's much appreciated. FWY, I really love your sliders. They're immensely useful.
Can I ask how you're generating the prompts for your example pics?...because while they're all impressive, most of the tags seem to have nothing at all to do with the image. Are they also AI-generated, or do you just throw a lot of random ideas at it and see what you get?
Its wildcards combined - https://civitai.com/models/98622/clonecleaner-wildcards-increase-character-consistency / https://civitai.com/models/69651/sdprompt-prompts-from-a-huggingface-dataset
I also use https://civitai.com/models/99047/location-wilcards (all put into one file called bg.txt)
prompt: person, bg for many
sdprompt, bg for others
then cherrypicked
Two of the example images are just single word prompts as well.
@ntc Interesting, I have not tried using wildcards with A1111. Thanks very much.
Can you share how did you "train to remove the need for triggers"?
It's custom research I've been working on, not easily used atm.
@ntc Oh, that's a shame. If you ever decide the tech is ready to be shared i guess it will be a big hit!
I get a warning when combined with https://civitai.com/models/216190/lcmandturbomix-lora-only-12mb-8-step-sampling-effect-is-superior-to-using-lcm-or-turbo-alone
dare_captivating_notrigger.safetensors unmatched=6 matched=522
there are a few layers that comfyui and a111 disagree on naming conventions for
if "lora_unet_up_blocks_0_upsamplers_0_conv" in k:
k2 = k2.replace("lora_unet_up_blocks_0_upsamplers_0_conv", "diffusion_model_output_blocks_2_2_conv")
if "lora_unet_up_blocks_1_upsamplers_0_conv" in k:
k2 = k2.replace("lora_unet_up_blocks_1_upsamplers_0_conv", "diffusion_model_output_blocks_5_2_conv")
so both are in the weights. This might be the cause.