Dennis | Welcome to the Outcast's Restaurant!🍽️ - 🧪ILXLv2🧪

NSFW

DENNIS from WELCOME TO THE OUTCAST'S RESTAURANT!

🌟UNIQUE CORE TAGS🌟

ILXL~~~
☄️/🧪
POS:
denniswor-1male focus

NEG:
Black eyes - only if you've had enough of black eyes

You may or may not add "anime screencap" in either the positive prompt or negative prompt

💡Download the Safetensor and check on any Safetensor_reader file for other unique tags

💡Be sure to check out the Pinned Posts in the gallery for Additional Showcases and usage of the unique tags!

༺𓆩༒︎𓆪༻

⏳INTERESTING HOW TOs⏳

❓How to fully use the character name + underscores?

💬You can remove the character name + ___underscores___, to improve LoRA flexibility.

They are not there for simply a different looking activation triggers

❓I want the look of the character, how to, when they are no unique tags?

💬Play around with relevant words of the character's appearance. I've never omitted them in training data.

(for ex. a character is commonly seen wearing a coat, add 'coat', and see what happens)

❓Your LoRA is not working, how to work?

💬Check checkpoint if it is overly bias on a type of subject.

💬Check prompt weighting (order of your tags in a prompt0

💬Check LoRA weights / Strengths

💬Check Tag weights (individual weights/influence of a tag in your prompt)

If none of the skill checks works, you can privately message me here in CIVIT, or find a DaBerry profile in Discord, or ask me over at Patreon.

ੈ✩‧₊˚༺☆༻ੈ✩‧₊˚

🛠️RESOURCE CAPABILITY🛠️

⚙️Character Focused Resource = Easily Generate the Character in any SFW / NSFW scenes

⚙️Gender Swapping

⚙️Race changing

⚙️Aging

⚙️Animification

⚙️3Dfication

⚙️Realification

⚙️Monsterification

⚙️Cosplay

🔧Detailed tips on how to use any LoRA resources + LoRA stacking in my Exclusive Prompt Guide!🔧

ੈ✩‧₊˚༺☆༻ੈ✩‧₊˚

🔐RESOURCE AVAILABILITY🔐

📜You can download the resource on CIVITAI

📜You Can Generate with the resource on TENSORART!

📜Only Available in CIVIT and TENSOR unless personally shared

──── ୨୧ ──────── ୨୧ ────🌱──── ୨୧ ──────── ୨୧ ────

🚩

🚩🚩NOTE REGARDING REPRINTS( ˶°ㅁ°) !!

Ask for my permission if you wish to reprint or upload my models on sites I have no knowledge of. I would love it if my LORAs can only be downloaded through CIVIT. Thank you (˵ •̀ ᴗ - ˵ ) ✧

Description

Adjusted Noise offset to 0.7,

optimizer to AdamW6Bit,

Unet to 2e-5,

Text Encoder 1e-4,

Scheduler cycles 1,

ND 32

NA 16

"Anime Screencap" testing

So...from what I delulud, I gave more importance to the tag's strengths so that the style can be better bloomed with 'anime screencap' and hopefully make it easier for the character himself to show when using only 1-2 of his core tags.

My delulu on unet is that if it is higher than the text encoder then the character's face will have more chances to bleed on other characters and the elements used in the dataset would also be more alike in the output.

Problem with what the output generating here is that 'anime screencap' definitely learnt the style well, and definitely is able to have the black eyes (I did not tag his eyes properly XD). So when replacing the core tag for his eyes with the normal purple eyes, the style learnt by 'anime screencap' is applied on the normal purple eyes.

Tho, I THINK I MIGHT BE TIPTOING T_T, Cause ND/NA is not crazy. And ND being at 32, the style is definitely bound to be learnt and influence the normal tags in prompting T_T So...I dunno what's the diff :`>

Another factor on anime screencap learning properly is the image dataset being consistent to the anime so :////

I only got what the what Unet, Text encoder, and Offset means. Dennis's LoRA shall be a Guinea pig for further uwu testings on a "anime screencap consistent img dataset"

Next test will be reducing the ND/NA to what I'm used to, 16/16. but I'll be adjusting the Unet and Text Encoder.