the best REALISTIC models for Stable Diffusion

In this video, we're going over what I consider to be the best realistic models to use in Stable Diffusion.
Guides, tips and more: jamesbeltman.com/
epiCRealism: civitai.com/mo...
epiCRealism Helper: civitai.com/mo...
NMKD Faces: huggingface.co...
NMKD Superscale: huggingface.co...
majicMix: civitai.com/mo...
badhandv4: civitai.com/mo...
ng_deepnegative: civitai.com/mo...
Analog Madness: civitai.com/mo...
How to install Stable Diffusion: • Creating (free) AI Art...
Stable Diffusion is an AI tool that allows you to generate high-quality artwork. Similar to Midjourney and others, It combines the power of artificial intelligence with your creative vision to produce stunning visual compositions.
music: massobeats - honey jam • massobeats - honey jam...
massobeats - waiting • massobeats - waiting (...

[@JamesBeltman]

thanks for watching - please let me know of any and all feedback you may have!

[@DejayClayton]

Thanks for the overview! FYI you have epiCRealism selected while explaining what settings to use for majicMix.

[@freespeech515]

More of STEALING someone Art style

[@JamesBeltman]

lol

[@freespeech515]

@@JamesBeltman AI is basically stealing art style and code. That is most innocent part of AI What is dangerous is this. They are training AI into replace human on repetitive tasks. Who AI helps? corporation. They may not need human labour once they train robots with AI All warehouse labour work in canada gone. There is no industry in canada. All industries went to china. All AI channels are propagating this instead of slowing down. We are tighteninng noose on ourself. AI wont be terminator style robots. FIRST victim is people who does manual labour. They will need less and less humans once robots start talking like humans. This only enrich big corporations. THIS MUST BE STOPPED at any cost. DEEP state fund this project through proxy. Every day AI THIS and that. AI is basically stealing things via training models.

[@ゴアに行く]

finally someone who makes quality videos ✌🏻

[@_ShaDynasty]

Haha agree. Most of them don't explain anything

[@UmarKhaleeq]

I never comment on KZbin, but I have to say… this was one of the most valuable and fresh videos I’ve seen in my years learning about growing as a creator. I was bummed out when it was over. Awesome job!

[@klaurcschwackerberg1880]

So you never comment on KZbin? I don't believe your crap since today I see you post the same comment on many other video's! You are SOOOO busted ! lol

[@takyc7883]

I have to admit, that last photo you made wwas far better than almost all the ai art i've ever seen!

[@Hellosirrrr]

I ve just found your channel and as data science engineering I’m in love with your content. I would love you to make also videos about Comfy UI and how to create right pipelines and mixing different models..❤❤❤

[@JamesBeltman]

ComfyUI is a bit of a headache when starting out - I'll definitely make a tutorial in the near future.

[@Philson]

majicMix is currently my favorite!

[@c.d.osajotiamaraca3382]

Excellent, you're a very good teacher. No fat all bones. Q: can you upload existing portraits into Analog Madness instead of an all text prompt? It might be helpful when generating an avatar and then throwing it back into AM to get a more realistic persona, or visa versa.

[@LexChan]

your video is awesome. just what im looking for. do more of these realism type of AI. more on controlnet, changes of clothing etc..

[@Minami1317]

thanks for sharing, I try to use all kinds of models.

[@rashshawn779]

Fantastic walkthrough!

[@malcolmreynolds4099]

Yay, I see a lot of my samples there during scrolling sections :)

[@calvinnguyen1451]

Awesome summary.🎉

[@LeeDong-w5q]

Thank you! Always nice to learn new things. Personally I'm playing with DevlishPhotoRealism SDXL now. I noticed an "Off" button between "width" and "batch count". Is it an extension or something else?

[@Rafaga777]

Very helpful. Much appreciated.

[@pastuh]

I was testing with HD images (each with a maximum height of 2000 pixels). I avoided using close-up shots and enabled the 'bucket' method. As a result, all the images turned out to be very clear and sharp. When attempting to merge with epicRealism, I noticed that my model's faces became more rounded and appeared unnatural. So epic, not so epic.. (at this moment)

[@NobukotoTV]

Thanks for your content, new sub in here. ❤

[@baco4632]

Really thank you this is best

[@KammaKhazi]

excellent content thank you

[@ekinshuya]

amazing I will give it a try

[@ssdkkkk7114]

Nice work, thanks

[@AuneJay]

Great Video ! lots of value, quick question... how do I install the upscale files ? or get them to show up in stable diffusion. Thanks keep up the great content

[@JamesBeltman]

in the video I used the Hi-res fix upscaler. You install them by placing them in your: stable-diffusion-webui/models/esrgan folder

[@hilaindie]

Nice Job

[@Ed-ym4tu]

I didn't have great luck with these settings. I think it comes down to a combination of the right model for the right style. I've gotten some great results, but it's been random and hard to reproduce.

[@JonatanE]

epiCrealism is great for a lot of things, but just like every other model there is, it fails completely att creating regular people, everyone is a super model. Except for Humans 1.0, which is my go-to Model because it generates actual regular people and has very high fidelity as well.

[@JamesBeltman]

You are 100% right - Humans is fantastic.

[@lespretend]

have you tried specifying the type of person you want? say like 'average 30 year old female' or even '3 out of 10 on beauty scale'

[@thanhdat-funnydogs]

love it.

[@FudduSawal]

can you please make an updated video for FIXING hands? what is the best way to get FULL BODY character without any badd hands and feets. Thanks in advance.

[@CapApollo]

fantastic content.

[@JamesBeltman]

Thanks

[@FlowerPower3000]

Very thank)

[@Dira-Han]

Superr...

[@benzpinto]

are these settings available in comfyUI as well? upscaler and denoising strength?

[@joglloq1673]

Can you tell me how install upscaler pls ?

[@felipenl6753]

Very good video. I have a basic question, in my SD webuai 1111 do not appear all the options for upscaler. Instead having the options showed in your video (Hires step, denoising strenght, upscale by, and 2 other marked in grey) in my webua appears only the options (Upscaler 1, Upscaler 2, Upscaler 2 visibility, Scale by Resize, Scale to). Could you please help me to find how to make these options appear in my webui and can practice with your video?. Thank you

[@3dtechnik]

I had the same issue. The trick is to download the *.pth files into the following directory: "stable-diffusion-webui/models/ESRGAN"

[@JamesBeltman]

@@3dtechnik Sorry maybe should of made more clear!

[@rafa-lk6lf]

@@3dtechnik didnt get it

[@JohnDlugosz]

4:40 Euler is pronounced "oiler".

[@JamesBeltman]

fair enough

[@ngbs75]

may i know what is the basic hardware requirement graphic card to get your AI generated graphic quality?

[@TomasGabriel83]

Where can I learn how to create realistic Ai art like you? I'm struggling to find courses on this...I've managed to create pictures but it's from software that are fully automatic (text to image), however I need to have more control. Could be paid courses, I don't care about that. I just want everything in one place. Any recommendations?

[@JamesBeltman]

It’s funny you say that because I’ve got a paid course on my website for exactly that.

[@TomasGabriel83]

@@JamesBeltman Then it's very well hidden. I was actually looking at your website before posting this comment. I saw no paid courses on there. All it says is "read more" on all the blog posts. There's not even a CTA. Or a course on there.

[@birdyorlando4822]

Can make a video for invoke ai models similar to this one?

[@ipurpleyou55555]

Жаль из России нельзя купить ваше руководство на сайте..

[@ThatHobbyHunter]

Thank you very much. A +1 Sub for you good sir

[@civivva4501]

I am trying to create an AI picture for a product page of a woman holding my product. But I don't know what prompt to use to put her arms in the right position. Any tips?

[@JamesBeltman]

Controlnet would likely serve you well. You could create a model using something like PoseMyArt and then import that into A111

[@Skystunt123]

what's the generation data for the selfie on the moon photo ?

[@elleeden2024]

I watched your video and still have NO IDEA how to create a picture? How did you get into "Automatic 11"?

[@JamesBeltman]

You need to download it, I'll be uploading a sometime this week.

[@elleeden2024]

@@JamesBeltman You will be uploading what this week?

[@JamesBeltman]

@@elleeden2024 A tutorial on installing it - my bad

[@elleeden2024]

Hi James, Do you know how to create a video with the speaker looking like a professional speaker, but being AI, for free? Thanks :)

[@JamesBeltman]

I don't really know what you mean. Do you want a video of a person speaking?

[@elleeden2024]

@@JamesBeltman Here is an example of what I mean, and AI presenter...search youtube for "AI Video GENERATOR : Create FACELESS KZbin Channel Using AI" Can you make a tutorial for that please and if it can be done free or only with a cost?

[@angeldelossantos2792]

how can i save my settings? every time i open it, it goes to default.

[@JamesBeltman]

I think there's an extension that allows you to do that but without it you can't

[@dianasharon5647]

Question!! Someone please help!! I downloaded the 8x_NMKD-Superscale_150000_G.pth, which folder do I put this in???

[@JamesBeltman]

stable-diffusion-webui/models/ESRGAN

[@Robstercraw]

I was looking for 8x_NMKD-Superscale_150000_G and could only find the pth file. what do I do with that?

[@JamesBeltman]

That's exactly all you need - you place in your stable-diffusion-webui/models/esrgan folder

[@Robstercraw]

@@JamesBeltman ah ok thanks😁!

[@user-zq6kh4gf3z]

Dude I bought your guide on making realistic images of models but i never received it...

[@JamesBeltman]

It should be instantanious. Try logging in again.

[@JamesBeltman]

Does it work now?

[@dr1nkm1lk]

same issue for me :( @@JamesBeltman is there way to contact you? I have a question

[@ekenmaice1435]

In my opinion, nothing beats the BRA(Beautiful Realistic Asians) model. Pair that with a binclass Lora and you have a realistic machine.

[@JamesBeltman]

It certainly is good, however, like MajicMix it’s too niche. Though, I don’t think I’ve ever heard of a binclass LoRa? What is it?

[@ekenmaice1435]

​@@JamesBeltmanyeah I see what you mean, I've really only generated Asian faces with model. Binclass is generally used to get 'better bodies'. But it's great because it adds grain and other imperfections to the image that make it look like it it was taken with a smartphone rather the professional photograph look that epicRealism or Realistic vision gives. I like to toss a good amount safe words in the Negative prompts to avoid nsfw images.

[@JamesBeltman]

That sounds very interesting. I really like generating “imperfect” but still amazing looking images. Could you recommend a Binclass LoRa? Which do you use?

[@ekenmaice1435]

@@JamesBeltman no that's just it. That's the name 'binclass' Lora.

[@ekenmaice1435]

I see epicRealism just got updated. And it seems this new version is really good. Would be really fun to try out. Realistic Vision 5.1 just got out also, I'd like to see how it stacks up against epicRealism.

[@shogunaa6498]

Hello, what kind of GPU and CPU is required for this kind of work? For a GPU, Will rtx 3060 12gb be enough or a 3070 8gb be a better one?

[@huyked]

yes

[@flickwtchr]

@@huyked There was an "or" in the question.

[@lucas1130]

Go with 3060 12GB, when it comes to AI, the more VRAM the better.

[@luke-ev5rz]

is this an app bro? im on the website but cant figure out how to use it!

[@JamesBeltman]

What website? It is an app I'll make a tutorial on installing it soon.

[@VortexVisuals]

Great info and content. But background music is anoying, and hard to hear and understand what you are saying ...😢

[@JamesBeltman]

I'll turn it down next time - thanks!

[@ysb321]

Which a1111 Theme you are using bro

[@JamesBeltman]

default bru

[@ysb321]

@@JamesBeltman but why slider look different than mine

[@Skystunt123]

he's using dark mode

[@TheSolsboer]

3 fingers is a marker that AI comes from the aliens

[@ronbere]

REv5.0 wins the game

[@JamesBeltman]

It's pretty good !

[@JarppaGuru]

programmed todo no ai. it wont drive car bcoz not peogrammed todo. so not ai

[@ATLJB86]

So many things missing here. Many good models like photon are missing. Prompt structure is missing. Camera angles and settings are missing. A lot is missing and too much to type. Simple advice I can give is use the real skin Lora by polyhedron so your photo won’t look like a digital painting. That’s the main problem with these models. They make real look too real! Nobody’s skin is that perfect. Second, start your prompt with a photo of. That will put you on the right base by default.

[@JamesBeltman]

Well yeah it's only a 9 minute video. I couldn't include every realistic model - I just included what I considered to be the best at what they do. What do you mean by: "Second, start your prompt with a photo of. That will put you on the right base by default." ?

[@ATLJB86]

@@JamesBeltman Oh ok, I got the impression that you only like the ones you showed. On the second part, I’m talking about the way models are trained. I’m speaking to the structure of the prompt. By saying A photo of a beautiful woman, it is already trained on what a photo is. That’s how you get the realism. If you put any photo into an image to text program, it will start with a photo of. It’s the universal language the AI understands. Also, you don’t want to use things like hyper realistic or hyper real, anything like that. That’s more of the CG look. The type of realism you get from something like Dreamshaper. You want to use Realistic or Photo Realistic. If you ever train a model, you will understand what I’m saying better. Just trust me 😁

[@ATLJB86]

@@JamesBeltman This is my prompt struct for photos: [STYLE OF PHOTO] photo of a [SUBJECT], [IMPORTANT FEATURE], [MORE DETAILS], [POSE OR ACTION], [FRAMING], [SETTING/BACKGROUND], [LIGHTING], [CAMERA ANGLE], [CAMERA PROPERTIES], in style of [PHOTOGRAPHER]

[@ATLJB86]

@@JamesBeltman so the prompt might look like this: pictorialist style, photo of a Cheerful girl, arms crossed, close up on face, the sun is setting over a rocky river, anamorphic lens flare, high dynamic range, overcast lighting, dutch angle, Kodak Vision3 IMAX, soft focus, Detailed,Best Quality,Luxurious

[@elodie2404]

You should specify in the title you're talking about a model only for SDXL, it is VERY HEAVY on SD1.5 and SDXL is much heavier than SD1.5... Don't just write "for Stable Diffusion" as if it is for every SD user ! Not everyone has a super powered PC to use SDXL

[@DaaronIrwing]

AI can't replicate same image from different angles that what you should now about it

[@JamesBeltman]

Elaborate

[@eugenenovikov671]

all these models paints very good only faces, but very badly woman in clothes with full body height

[@AJ_real]

Ultra realistic AI bothers me, so much potential for fakes.

[@campcrafter4613]

AI?

[@jcdeton72]

I am so tired of seeing AI-girls. Its the most boring and overused thing you can do with AI. These models are really awesome and they can do a lot more. Like awesome scenery or fantasy creatures.

[@JamesBeltman]

I agree. But I also understand that 95% of users are mainly interested in generating fake women therefore I showed off how each model can generate fake women.

[@BVLVI]

You complained about the ethnicity of the models the entire time dude. Look at the sample images if you don't like them. Don't use it.

[@JamesBeltman]

I evaluated it's strengths and weaknesses. Good at lighting, bad for anything but asian woman.

[@TheDocPixel]

@@JamesBeltman I also add asian and all the Eastern nationalities as a negative style. Don't forget to include NSFW and child to it too, because most of the models on Civitai are for p***

[@crypty9365]

openly racist video?

[@carmillasmith7743]

Where can I go to find a community of professional AI artists?

[@Zeironn]

in my stable diffusion theres no option for upscales. what do i do wrong ?

[@Zeironn]

theres also no controlNet. how do I get that