extremely detailed (no trigger) - sliders.ntcai.xyz

extremely detailed (no trigger) - sliders.ntcai.xyz - v2.0

NSFW

v2.0 Fixed negative. Range -1 to 1. No trigger required
Fork this and train your own slider at https://sliders.ntcai.xyz/sliders/app/loras/e3d5252c-5e6c-49e3-bd19-5ec49998fdc2

Make everything more detailed, no trigger required. Examples at 1 strength.

Trained for 6000 instances with sliders

batch size 12, 500 steps, 1.5 hr on 1x a6000

Trained on https://civarchive.com/models/215796/top-rated-turboxllcm-super-mario-merge

Support my work on patreon and get access to over 1000 unseen loras with many new ones daily. Plus, easily train your own, no hardware needed (coming soon)! https://patreon.com/NTCAI

Any proceeds go to funding my compute addiction for more and better finetunes.

Description

FAQ

Comments (40)

yofoton174609Mar 10, 2024· 3 reactions

CivitAI

thanks for doing it without a trigger! Awesome! This is perfect for creating those small flakes, such as raindrops, snow fall, and similar microparticles. Perhaps a better name would be "Particle Enhancer"

ntc

Author

Mar 10, 2024· 1 reaction

It does go a little crazy with the particle effects. I like to name these based on the positive trigger words so I don't have to think too hard but if you want you can fork it and create your own / release it. It does take a patreon membership but I provide the hardware for members to train on.

storm7900gs425Mar 16, 2024· 3 reactions

CivitAI

wonderful! This is the first time I have encountered such a helpful Lora!

kurodahuga843Mar 17, 2024· 6 reactions

CivitAI

Miscellaneous (speculation)

IMPORTANT: May not work well for ANIME type, 2.5D is also suspect, so 3D and REALITY type is recommended.

Detailing for characters only. It does not affect the background much. However, it seems to react if inorganic objects such as cups are the main subject. (I think Gallery's is a wonderfully decorated cup.)

May not be applicable to animals. Dogs and cats could work well if they were clothed.

To the author.

I didn't submit this because it's not a comparison chart, but I'm glad I applied it to WAIFU and it's so beautiful that I fell in love with it. It was very good. I appreciate it.

windlike_gustApr 2, 2024· 1 reaction

CivitAI

The first version works fine in ComfyUI. The second one gives errors:

lora key not loaded: diffusion_model_output_blocks_2_2_conv.alpha

lora key not loaded: diffusion_model_output_blocks_2_2_conv.lora_down.weight

lora key not loaded: diffusion_model_output_blocks_2_2_conv.lora_up.weight

lora key not loaded: diffusion_model_output_blocks_5_2_conv.alpha

lora key not loaded: diffusion_model_output_blocks_5_2_conv.lora_down.weight

lora key not loaded: diffusion_model_output_blocks_5_2_conv.lora_up.weight
ComfyUI: 2094[130646](2024-03-31)

windlike_gustApr 2, 2024

I just found this: "If you are using hugging face's script to train your own lora, there is a file "https://github.com/huggingface/diffusers/blob/main/scripts/convert_diffusers_sdxl_lora_to_webui.py" which can convert the lora model to the form that fit with webui&comfyui".
source: https://github.com/comfyanonymous/ComfyUI/issues/2689

ntc

Author

Apr 3, 2024· 2 reactions

@windlike_gust Hey yeah theres disagreement on the names of those layers between sdwebui and comfyui so I duplicate the weights for both keys.
code:
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")

output_tensor[k2]=tensor_map[0].get_tensor(k)

sleepyroboApr 3, 2024

oh wow I was just looking into this myself, it does indeed appear to be the LORA, and your saying the fix is to use to conversion script? hmm ok

ntc

Author

Apr 5, 2024

@sleepyrobo yeah that's teh best compromise I found. It has these warnings on comfyui and a1111 though.

jirehliston420Apr 7, 2024· 3 reactions

@ntc This will cause the model to fail under diffusers.
because of this code in diffusers/loaders/lora.py

# First, ensure that the checkpoint is a compatible one and can be successfully loaded. state_dict, network_alphas = self.lora_state_dict( pretrained_model_name_or_path_or_dict, unet_config=self.unet.config, **kwargs, ) is_correct_format = all("lora" in key for key in state_dict.keys()) if not is_correct_format: raise ValueError("Invalid LoRA checkpoint.")

3645345636dzJun 23, 2024

I still don't get it. The conversion script doesn't finish, it outputs some errors :( Isn't it possible to upload the fixed converted version here? Would be so helpful for stupid guys like me...

ShivaeApr 7, 2024· 5 reactions

CivitAI

The quality of your lora is fantastic!

NeoCorAIApr 12, 2024· 1 reaction

CivitAI

I use Easy Diffusion with Pony V6 and it doesn't work, in Civitai with the same model it works.

Any help?

ntc

Author

Apr 12, 2024

How does it fail?

NeoCorAIApr 12, 2024

@ntc Could not load the lora model! Reason: Sorry, could not load X:\Diffusion\models\lora\detailed_notrigger.safetensors. You're trying to use a SD1 LoRA model with a SDXL Stable Diffusion model. They're not compatible, please use a compatible model!

ntc

Author

Apr 12, 2024

@NeoCoR ha its not an sd1 lora. Perhaps this has to do with using diffusers on their backend?
I can't solve this easily on my end without guessing so can you reach out to them? Technically, it might have to do with duplicate weight names for some layers added for compatibility (see the other comments)

CreamDisperserApr 22, 2024· 1 reaction

CivitAI

Hi, this may be a foolish question. When I tried loading the LoRA (using A1111 WebUI for SD Forge), I get the following message:

Loading detailed_notrigger.safetensors for SDXL with unmatched keys.

Is this a problem, an error, warning or something else? I guess ultimately, do I need to install any other extensions for the LoRA to work?

Thanks in advance!

ntc

Author

Apr 22, 2024

Hey, safe to ignore. There are duplicate weights where a1111 and comfy disagreed on name

CreamDisperserApr 22, 2024

@ntc Thank you for the quick reply :-) Appreciate it my man!

mayan007Apr 25, 2024· 7 reactions

CivitAI

It makes even realistic images into 3D looking cgi like. Any fix for this ? Am i using some wrong setting ? It increases details nicely though !

ntc

Author

Apr 25, 2024

Maybe you can train a new one with 'extremely detailed photo'. Code is at https://github.com/rohitgandikota/sliders or you can use my app/hardware for $5 a month.

mayan007Apr 25, 2024

@ntc Link to your app/hardware ?

ntc

Author

Apr 25, 2024

@mayan007 https://sliders.ntcai.xyz/ sign in with patreon https://www.patreon.com/NTCAI

MarkotrivMay 19, 2024· 3 reactions

CivitAI

Great quality

GreatHuntMay 30, 2024· 6 reactions

CivitAI

great lora sometimes I wish it impacted the face shape less.

StreamTabulousJun 22, 2024· 1 reaction

CivitAI

is there a version 3 coming you legend

ntc

Author

Jun 26, 2024· 6 reactions

Yeah! Version 3 will have DoRA for better quality, native pony support and sd3-medium. Coming "Soon"

LifeToggleSep 26, 2024· 1 reaction

@ntc Love you're work. I would like to know, once version 3 is available, will it show up on this page? i just want to make sure i dont miss it. Thank you for all of your work.

GipnoJun 22, 2024· 13 reactions

CivitAI

How can you tell if this LORA is working? How to trigger it?

QH96Jul 5, 2024· 5 reactions

do the same prompt with the same seed, with and without this lora and compare the difference

Gr1efJun 5, 2025

It might work better with some models than others. It's based on SDXL 1.0

GipnoJun 26, 2024· 46 reactions

CivitAI

Well, the author does not answer the question. It is not clear how to make LORA work and trigger it. I'll look for other analogues.

ntc

Author

Jun 26, 2024· 14 reactions

hi, you add the lora to your prompt without trigger words. It works better on SDXL types than pony types.
Sorry for the delay I am working to integrate sd3-medium / get next versions out (+ client commitment to pay for more compute).

hentaiccdmJun 23, 2025· 3 reactions

extremely detailed (no trigger) - I do not understand the question -- the title says - NO TRIGGER --

DarkCobraJul 8, 2024· 18 reactions

CivitAI

I use this on Pony models and it works okay (i know you say its better for sdxl base and maybe it is but it still works too on pony). Do you think you would make a pony version in the future? I literally add this to all my prompts now as its very useful and i like you can make it subtle or strong depending on your needs.

ntc

Author

Jul 8, 2024· 11 reactions

Thanks! Pony LoRA is training. It's in progress, I hope to upload it soon
https://sliders.ntcai.xyz/sliders/app/loras/5001c332f87144fb9f61a6815e6faa17

acgc99Jul 31, 2024· 1 reaction

@ntc Will you upload it to Civitai? Or that link is a WIP?

DarkAgentAug 11, 2024· 1 reaction

would like to see a pony version as well

4703559Jul 17, 2024· 17 reactions

CivitAI

Great lora. It seems to play very rough with DPM ++ and works very well on lower steps with euler based models, which is interesting.

DabGuyJonesAug 28, 2024· 22 reactions