ALL IN ONE <3 // (Skin – Hands – Eyes) by Polyhedron

ALL IN ONE <3 // (Skin – Hands – Eyes) by Polyhedron - v1.0

NSFW

<3 Please join my PATREON page, where you get additional content (for SDXL) and personal help, IF I can help you^^:
https://www.patreon.com/polyhedron_ai

*Publishing Notes:

Version 1.0 is a new trained LoRA at the edge of what my computer is capable of with a 3060 Ti 8 GB VRAM: containing a training with 355(!) high-res pictures trained with 35500 sampling steps. It works. I guess. :D Still a blackbox. Use it wisely & Enjoy.

@Hands: It could work, see my gallery below. But it's hard and tricky. But maybe it helps anyway ... :)

Description

FAQ

Comments (12)

SnG17Aug 6, 2023· 6 reactions

CivitAI

Bro you using too many negative prompts and Embedding. idk how to evaluate the Lora. xd

Edit: nvm not all of them, just the hand example. xd

ok, it's overcooked, it breaks at 0.6 above and it works at 0.3 and below.

Polyhedron_AI

Author

Aug 6, 2023· 1 reaction

Yes, it only works with smaler weights. :) Some day I will find out how to make it perfect. But not today. :D

SnG17Aug 15, 2023· 2 reactions

@Polyhedron_AI nice. I'll wait that day.

hugitusioqiqere762Aug 7, 2023

CivitAI

Hello, great Lora one of the best i could say. Thank you for that. Is it possible for you to tell us more about the training process itself? Like what are you using to train the Lora with what kind of settings, what kind of images you were using, and how you tag your images. That would be great :)

JaradisSep 5, 2023· 1 reaction

CivitAI

I had a very hard time getting this to work. Your skin lora works well, but this one completely drown out all my other prompts and had quite a lot of mutilations no matter what strength I used. My subject was in a field, did dozens of images just fine, but once I added this lora it kept putting them inside a bedroom, laying down, and just ignoring all my other prompts.

Polyhedron_AI

Author

Sep 5, 2023

Perhaps you can share your prompt with me, so I can check it out ... :)

JaradisSep 6, 2023

I switched to SD 1.6 today, so it's hard to get the exact same thing again and I don't remember exactly which items I was working on when it happened. But if you take the example here:

1girl, caucasian, white girl 25 years old, fairy, thin, tiny, petite, medium breasts, (two pair of symmetrical red fairy wings:1.3), (large wings:1.2), beautiful, perfect face, long messy hair, necklace, , small jeweled princess crown, (red hair), dnd, lake, field with wild flowers, mountains, fantasy, (standing|walking|dancing), (masterpiece, photorealistic), dappled sunlight, studio lighting, (photorealistic:1.2), <lora:more_details:0.8>, dslr, ultra quality, sharp focus, tack sharp, dof, film grain, Fujifilm XT3, crystal clear, 8K UHD, highly detailed glossy eyes, high detailed skin, skin pores, <lora:all_eyes_hands_skin_fin:0.1>
Negative prompt: asian, nude, topless, nsfw, painting, art, cartoon, anime, b&w, (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, logo, signature, watermark,
Steps: 28, Sampler: DPM++ 2M SDE Karras, CFG scale: 7, Seed: 3239617382, Face restoration: CodeFormer, Size: 912x512, Model hash: 65f28f91e0, Model: realisticFantasy_v20, Denoising strength: 0.3, Hires upscale: 1.5, Hires steps: 15, Hires upscaler: 4x-UltraSharp, Lora hashes: "more_details: 3b8aa1d351ef, all_eyes_hands_skin_fin: b37c3e9a2db1", Version: v1.6.0

I use both the Realistic Fantasy v2.0 and the Realistic Vision v5.1, used Realistic Fantasy for the images below:

Without your Lora I get: https://i.imgur.com/sGci2Fm.png, with <lora:all_eyes_hands_skin_fin:0.5> I get: https://i.imgur.com/TXvGdsT.png, at 0.3 I get https://i.imgur.com/QZwDDIv.png. I have to go down to 0.1 just to get back to my requested area: https://i.imgur.com/U0Gh0VX.png It's a good image but at 0.1 I don't see any improvements.

But basically if I am set at 0.3 or higher I can never change the background from what it shows, not really. It's like that Lora is just overpowering everything.

Polyhedron_AI

Author

Sep 7, 2023· 1 reaction

@Jaradis So, I tried this and I guess, it works for you:

1girl, caucasian, white girl 25 years old, fairy, thin, tiny, petite, medium breasts, (two pair of symmetrical red fairy wings:1.3), (large wings:1.2), beautiful, perfect face, long messy hair, necklace, , small jeweled princess crown, (red hair), dnd, (lake, field with wild flowers, mountains, fantasy:1.5) (standing|walking|dancing), (masterpiece, photorealistic), dappled sunlight, studio lighting, (photorealistic:1.2), <lora:more_details:0.8>, dslr, ultra quality, sharp focus, tack sharp, dof, film grain, Fujifilm XT3, crystal clear, 8K UHD, highly detailed glossy eyes, high detailed skin, skin pores, <lora:all_eyes_hands_skin_fin:0.3>

Sampling steps: 50
CFG Scale: 4.5
New Sampler: DPM++ 3M SDE Exponential

This Lora is totally overtrained. :D So, keep the impact low between 0.1 to 0.3. It's enough to enhance the skin textures, helping with eyes & hands. Cheers! <3

chzbroSep 9, 2023· 3 reactions

@Polyhedron_AI Personally, I want my lora overtrained. It bugs me that more people don't do it. Keeping the weight low means that you can use more lora without scorching your image. Often I want to use 3 or 4 lora on the same image, which is tough to impossible when they're all weighted at 1.

Ninjun7Mar 14, 2025

CivitAI

It made me lose control of my Prompt.

Unfortunately, that wasn't good for me! For me, Lora should work as an aid and not as a command.
Eu agradeço o compartilhamento. Even so, I gave the post a positive rating.

Polyhedron_AI

Author

Mar 14, 2025· 1 reaction

Reduse the LoRA weight to 0.3 or lower and try again ... :) But yes: It changes your original picture, I hope for the better ... :D Thanks!

Ninjun7Mar 15, 2025· 1 reaction

@Polyhedron_AI Thank you for your attention and help.

I've already tried this, I got average results with 0.2. It's changing the result a lot. I'll use it when I really need HANDS to appear.
Excuse my bad English.

LORA

SD 1.5

by Polyhedron_AI

Download (Beta) View on CivitAI