REAL SKIN Lora - make your AI Images look like real Photos!
ฝัง
- เผยแพร่เมื่อ 22 พ.ค. 2024
- This Lora can help you get Skin that looks real. This Lora adds blemishes, wrinkels, pores and much more details to your AI images. Stop having your AI images look like digital art. Make them look like real photos now for free.
#### Links from the Video ####
Skin Lora: civitai.com/models/109043?mod...
Eyes Lora: huggingface.co/casque/detaile...
LowRA: civitai.com/models/48139?mode...
Ultrasharp Upscaler (search the page for 4x ultrasharp): upscale.wiki/wiki/Model_Database - แนวปฏิบัติและการใช้ชีวิต
hey i am Italian , the best video with the most technical features ever seen ! is long time i was looking for real skin tutorial !!super professional !! 👏🤩
Hi Olivio, thank you for your videos, it's always a pleasure to learn new things watching your content!
instead manually inpainting using img2img, you could use that eye enhancer lora thing on aDetailer (using mediapipe_face_mesh_eyes_only as detection / inpainting models), that extension will do inpaint for eyes, and then there's a textbox for detailed_eye-10 lora can be put.
Thanks Olivio, you reminded me that I'd downloaded this but forgot to use it.
DPM++ 2M SDE Karras is my favourite and go to sampler.
I went from multiple different models to DeLiberate to ReLiberate to Photon.
Photon is absolutely outstanding. I am so glad you recommended it.
Also, I enabled the setting to save both the image WITH and WITHOUT Face Restore.
I can always try taking the face restore output and cut/pasting parts of the face to the non-face-restore image and inpaint to match if I want to.
Realistic mixes and what's achievable are getting more amazing all the time. IMO edge of realism is still worthy, and epiCRealism PE v3 (v3.5 now too) is a new find that's superb, as well as a number of other excellent new models and updates recently. I find hive moderation interesting as a gauge to see what another AI thinks of images; historically it's labelled realistic images for me as 99.9% SD pretty much without fail, but with recent images on newer models I've had gens that it thinks could be or even almost certainly are MJ, including one that it branded as only 0.1% likely to be AI, which I never expected to achieve. I now never use face restore as ADetailer works well and doesn't mess up the integrity of image noise or film grain. The CFG fixes that the Dynamic Thresholding extension gives I find are worthwhile. I've not come across sampler options such as discarding the next-to-last sigma or applying I2I color correction being discussed, perhaps a bit too esoteric and buried in A1111 settings, but they can be worth playing with too.
Photon is one of my favorite models also 😊 I’ve been using it for quite some time.
Awesome job my bald genius! Thanks!
Everything about these are perfect, except one imperfection and that is the pupils.
something that has really upped my photorealism, in regards to skin detail, is adding 'natural contour aesthetics, sweaty meat' to the positive. I know, I was apprehensive about the 'sweaty meat' prompt too, but it adds really great detail to the skin and almost a natural skin oil/highlight.
Lol
Sweaty meat 😅 that beats my ((skin pores))
Will wait for the safetensors version of the skin lora.
The issue I tend to have with these loras is that they generally change other aspects of the caracter, clothing and composition. I imagine one could do img2img, but i don't normally get great results that way.
You've misspelled portrait (portait) btw. I've noticed you occasionally have misspellings in your prompts, and I wonder how they are affecting your results.
getting extremely realistic skin with epicrealism (without lora), a checkpoint i can only recommend.
only when upscaling, lots of skin detail is lost....
will try it out tho. thanks for your videos Olivio. 🤗
1) Small details, skin texture is smoothed and disappears due to upscaler (model) when scaling (in img2img, hires fix or Extra).
I tried different upscaler models - 4x_fuzzyBox preserves the skin texture the best. All the others smooth out the skin too much.
2) Also texture is strongly affected by sampler - for example Euler smooths and blurs skin texture, so it is better Not to use it when scaling or on the second pass at hires fix.
Thank you very much bro🙏
For realistic skin I'm often using the Analog Diffusion Lora, but I think the checkpoint you use has the biggest impact on the quality of the skin anyway.
There's a model called "Am I Real" that's also very decent
great 🤩🤩
Just tried it with creatures and it’s fire
kinda curious now, if you were to increase the skin lora weight to a rediculous amount and then give it a horror prompt
What software is needed to transform A.I. images into ones that look like this?
Is there an alternative to LOWRa? It seems like it's no longer available... :(
By the way, to enable previews/thumbnails of your Loras/embeddings, download an example image of the Lora you want the thumbnail of, then save it in the same folder with the same name as the Lora with the extension JPG (not JPEG)
You can literally do a render and click the icon on the preview image that’s missing and just set it to your custome image using the lora
@@cchance Oh, thanks for the tip!
@@HamedEmine NP ya it's pretty hidden and most people overlook it
the civitaihelper extension (downloadable from within A1111) will automatically get the picture from the models presentation page or leave it at your own defined preview pic,
it also lets you check for update versions of TI HYP CKPT LoRA LyCoris and can insert trigger words of LoRa/LyCoris with 1 mouseclick
.PNG files work as well.
I'm pretty sure Euler is pronounced "oy ler" after the Swiss guy who invented graph theory. not yoo ler how it looks.
It really depends on the language of pronouncer. We pronounce this surname as "A-ler" ("A" sound like first two letters in english word "aim") in my country.
I guess it will be a problem if I want to take multiple photos of the same character, it will have the marks somewhere else each time.
personally, i add into the prompt 'birthmarks, flushed face, blushing cheeks' as desired with additional modifiers. this gives it enough attention to go away from the porcelain skin (neat descriptor Olivio)
What are the additional modifiers?
ie. (blushing cheeks:1.2) @@KINGLIFERISM
How to deal with the color difference between the skin color after redrawing the mask and the original skin color of the character？I have been troubled by this problem for a long time！
when are you going to make a video about SDXL Lora?
Hi Olivio. Why is there no model available in the dropdown menu for selecting in the Control Net section? Where can I download the models related to Control Net from?
Hey Olivio, are you using dynamic cfg threshold yet? I find it to really help with contrast and skin imo.
Can you elaborate please?
I didn't really notice that the eyes look unnatural, but something that bothers me is that it looks like the teeth are growing out from the lip...
How do you use the same face for every rendering please?
오늘도 감사합니다~!
What's better overall "Loras"? or "LyCORIS"?
No safetensors format for this one.
06:00 - These aren't even an option in my settings. I don't see them.
This Skin Lora gives me a grey textured overlay over the whole image when used with Hires Fix... Tried with different 1.5 models, still bugged.
Can we generate 3d models to ai image generator ??.
There are still people that use restore face?
New users.. Everybody else upscales
Would you mind helping a newbie? Why can’t you use face restore with a Lora? Thanks a lot
@@SB-hj9ueFace restore will overwrite the face skin with its own, after the Lora has worked.
@@PippetWhippet thank you so much!
@@SB-hj9ue think of it as a relict of past times. You rarely have issues with faces with current models. Its hands, limbs, architecture and perspective artefacts that will eat your time.
How ato add Lora in to stable diffusion? I dont see a link to install or use ? Is it online or do I have to install stable diffusion to my pc for using Lora? Please guide. Thanks
I would also love to know how this can work with Roop!
foudn any solution? roop creating faces that are too perfect, and so looks unrealistic, freckles and blemishes dont transfer
Hello, can you make a video for restoring old and damaged photos?
You could get relatively decent results with just img2img with a low denoise setting
Nice. Do you have a video for stable diffusion on group of peoples? Number of person, different faces, asian faces with trained model, etc
Any tricks or links guys?
You can use adetailer extension or inpainting for that
@@OlivioSarikas thanks. Thats what I'm struggling with. 😅
Some embeddings have almost this same effect.
Hello Olivio. I tried to use this Lora, but when setting the value to 1 in the lower field of the prompt, I got a very sharpen broken image. I lowered the parameters to 0:3 and it got better. In the video I saw you put 1. I don't understand what the reason is.
Exactly what I got as well, double checked everything else.
pardon my ignorance,but what program A.I. are you using?
it's called A1111
What is the system configuration required to run stable diffusion webui ?
6gb of GPU VRam is recommended
where should i put the ultrasharp 4x file or folder ? is it in lora folder or ?
models/ESRGAN
Hey Olivio, why are you not using CivitaiHelper extension? Saves so much time. :)
Mine stopped updating long ago. Yours still works?
That's a great extension, my only issue is that it doesn't work with Lycoris. Hopefully version 2 (ModelInfo Helper) will solve that, unless I'm missing an update but I don't think so.
@@thanksfernuthin I believe it stopped updating because they're working on the new version I mention in my comment. My install still works for TIs and Loras.
@@Elwaves2925 Thanks. Mine hasn't worked in months however. PCs are so different! No way to know what's causing a problem sometimes.
@@Elwaves2925 Now that multiple people have mentioned it. I checked and while I always update my extensions, my webui says v1.2.1, but according to github there's already 1.4. This would explain why I don't have zoom controls for inpainting, but other than that, I've yet to notice a lacking feature (also there's a standalone extension for zoom btw). I guess I'll rather stay on 1.2.1 then, as Civitai helper is my most useful extension, the update + meta information grabbing is invaluable.
But you dont plug the eyes lora in prompt on the last step ??
I'm a complex and wonderful creature 😆😎
How to I make my face prettier using ai
I can't find the options in settings for turning on prompt inputs for hires fix, on 1.5
maybe you need to update you A1111
@@OlivioSarikas All updated, still not there. Strange.
@@jonescrusher1 did you put the settings i show in the video to turn them on?
@@OlivioSarikas EDIT: Updated A1111 did work for me.
This is very nice, but I guess you can't really use ROOP with these as ROOP will overwrite a lot of the work. Which then gave me this idea: what if you trained LoRA using images from ROOP & face restore instead of real images? That way you would only need one great photo of the subject to make what could potentially be a decent LoRA for that person?
And??? Did it work? If so how?
It was just an idea. I haven't tried it.@@KINGLIFERISM
great idea, but wouldnt that be limiting due to the fact that at the base/source of that is just 1 face? and all the pictures with be be created using roop, so wouldnt the face be too perfect without any blemishes and non realistic?
@@Faded_Fragger I tried ipadaptor and reactor on top then upscale and it is hi res. BUT takes long and the result does look souless but high detail
@@KINGLIFERISM wdym souless? also, ill prob try this guy's idea on making a lora
Will this work with roop? 🤔
I'd love to know this too! How can you add this with roop?
Probably have to run image2image again with around 0.3 denoising after running roop face replace.
Euler is pronounced like "oiler".
File size is too large
#### Links from the Video ####
Skin Lora: civitai.com/models/109043?modelVersionId=117467
Eyes Lora: huggingface.co/casque/detailed_eye-10/tree/main
LowRA: civitai.com/models/48139?modelVersionId=63006
Ultrasharp Upscaler (search the page for 4x ultrasharp): upscale.wiki/wiki/Model_Database
.safetensor files exist because of the novelty of sharing these files between others. I won't put a malicious script in my own embedding, but can I be sure others won't?
Can you make a tutorial about OneTrainer? Thank you!
LowRA -> 404 :)
Nah, the point is to get perfect skin beauty.
people always want what they don't have 😅
How sensitive is this to spelling mistakes, e.g. "photoportait"?
Just wondering, what is the point of images like this?
Eehhhh... not what I'm looking for. Blemishes in general. Greater skin detail is great and prompting for that usually helps. Pores and skin cells. But I'm sure this is exactly what some people are looking for and that's cool.
I've been looking for something like this. Perfectly smooth skin is featureless, boring.
You want my advice friend? Change your major model to PHOTON.. that's the King of Photorealism. Except you're not going for that
@@3oxisprimus848 Well... as for me, the best tor photorealism is Realistic Vision 4, then Reliberate and on the third place is Photon. These were for persons. In architecture there are other models.
yet another 'PLUGIN' to get results....by the time you install and go to use it - a better one comes out...and you spend more time chasing the next plugin than doing any work. No point in doing that image - cuz next week Blah Blah plugin will solve all my issues.....etc etc etc
Any for any race besides white?
RealisticVision create realistic skin without this lora =)))
seems you like older women haha, just input that on prompt, problem solved
First
Still looks fake.