HOW TO CREATE PHOTOREALISTIC AI IMAGES | Stable Diffusion
Вставка
- Опубліковано 24 сер 2024
- A lot of people are struggling with generating AI art to their likings on a local machine, not Midjourney or DALL-E. Today, we go over a new model, some settings and a couple of different prompts to see what is possible with this model.
STOP BY THE DISCORD AND SHARE YOUR ART! / discord
Realistic Vision: civitai.com/mo...
Prompt: RAW, analog style, A stunning portrait of a young, beautiful woman with striking features. She should have long, flowing hair, a captivating gaze, and an alluring expression. The background should be neutral and simple, allowing the woman to be the image's focal point. Please pay close attention to the details, such as the highlights and shadows on her face and hair, to create a lifelike and realistic image. Experiment with different lighting techniques to enhance the mood and atmosphere of the portrait. The final product should be a masterpiece that captures the essence and beauty of the woman, ((highly detailed skin, skin details)), sharp focus, volumetric fog, 8k UHD, DSLR, high quality, film grain, Fujifilm XT3
Negative prompt: ((((big hands, un-detailed skin, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime)))), (((ugly mouth, ugly eyes, missing teeth, crooked teeth, close up, cropped, out of frame))), worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck
What's your email address
What app u use?
@@ikhwanfathur6181 Stable Diffusion
Why can't I download this?
@@BattleBrotherCasten you can😆
just want to say that you're the first video ever about stable diffusion that i watched and my interest peaked. been playing with chatgpt for a while now
My experience is that SD language skills are very limited compared to GPT and that long verbose descriptions are counter productive 99% of the time. What for worked best for me is to keep the prompt as short as possible, with a high densitiy of keywords and short phrases describing different concepts and then adding weights to them. If you don't do this, every keyword you write is a requirement that massivly restricts to creativity of the model and also softens the focus on the main subject. For example if you put something into the negative prompt like "additional hands" the AI may simply try to prevent creating anything with hands at all and you may miss a lot of interessting stuff. So one strategy is to leave the most important keywords and phrases at the default weight of 1 or set this a little higher (1.1 or 1.2) and the rest should be below 1. So it will be more a hint than a requirement. The same goes for the negative prompts. So a prompt would look like "(knight:1.0), (castle:0.5), (night:0.5)" and you can easily just increase or decrease a number here and there if you have bascially working prompt instead of having to reword it over and over again. "Filler" words are mostly ignored anyway though they change the output a little. I'd also only add things the the negative prompt if required, like add "blurry" only if your images are blurred and if the generated persons are really too ugly, I'd something like "pretty" to the positive prompt. There is no generic negative prompt one should have all the time. Redundant descriptions with alternative wording is also mostly less effective than just putting more weight into the things you want. Also I don't know if it's just your prompt and seed or if you are just seeing a variation of the guy used to train the model. I mean in the end not only the quality is relevant but also the creativity, like how often the AI comes up with intriguing backgrounds, situations and details. Anyway just my 2 cents. :)
Thanks for your advice! One thing I’m confused - if you don’t put “additional hands” in the negative prompt how do you prevent SD from generating extra hands?
@@justinnine4940 You can increase the Greg-factor. ;-) Seriously, if you have artists in your prompt, the inferring will be more focused on the works of these artists who ususally simply do not draw extra hands. ;-) So giving that artist more weight can help. If your results are then look lets say for example more like oil paintings and you don't want this, you can add "oil painting" to your negative prompt. Also it's not totally forbidden to use "additional hands" in your negative prompt but if you do, add a low weight: That was also my main recommendation, to work with weights instead of verbose prompts like "Please be so kind and draw my a wonderful picture of...." and similiar nonsense. Stable diffusion is not ChatGPT! So if you add "extra hands" to your negative prompt, it remains very debatable if the SD AI actually understands that, my very impression is definitly that it just reads "Do not draw hands!". For people with lots of time it might be interessting to research the change of the distribution, like how the amount of "no hands" changes compared to "too many hands" depending on changes in the negative negative prompt. Maybe it's also possible to specifically train the model about what "addtional hands" actually means. But it hasn't been trained for this, so how should it know? I mean there are most likely not many images in the original training set with prompts like "here we have a guy with extra hands" and "this is a picture of a guy who only has the normal amount of hands". ;-)
All that said, I got the impression that while the AI doesn't understand quantifiers, like what are "too many fingers" and such, it seems to understand to a certain degree negative prompts like "disconnected hands". Obviously to be combined with a weight below 1.
@@testales that makes perfect sense, thank you!
Now they're really good, long verbose makes the scene just right if you express yourself plainly.
@@DaniboyBR2 No, that's technically impossible. SD is a different thing in comparison to a 7B+ language model for example which alone requires a big data file and has been trained differently.
Wait, are you telling me these pictures were fully AI generated, and these two do not exist ????
yes, that's right! amazing isn't it :)
Of course, lol.
Correct
Thanks for the video. I agree that there are problems with identical faces. I will fix this problem in the near future.
Thought the profile pic looked familiar... thanks so much for commenting! Great work on RV 1.1 and 1.2, some of the best stuff out there!
Did they fixed that as well? Cause i was wondering if it's possible to create multiple images of the same face?
This is awesome! I am so surprised this way of writing prompt works so well, thanks for the tip
It doesn't.
I’m having so much fun with SD. Thanks for the video.
Hey, thanks for all your tips, these will definitely save myself some time as the trial and error process is extremely long when running this on a subpar system.
Also, just on a side note, CFG Scale doesn't mean Config Scale, it's an acronym for Classifier Free Guidance Scale, whatever that means.
I’ve just realized that I’ve been saying “config,” over and over again. It’s what it’s stood for in my brain for over a decade but thanks for the heads up haha! If you want to experiment more on your own system, you can set up a XY plot with loads of parameters and let it run overnight or while you’re doing other things.
Thank you so much! This has really improved my images. They look much more real now😊👍👍👍
This prompt looks like studio changes and fixes 😂
"Please pay close attention to the details, such as the highlights and shadows on her face and hair, to create a lifelike and realistic image. Experiment with different lighting techniques to enhance the mood and atmosphere of the portrait"
you can improve the prompt?, and share with us?
love the content, I’m binging your backlog and subbed of course. Please keep it up with the short-form tips and how-to, ideal for snacking!
The evolution of A.I as it is... 🤔Mind-blowing! 😮 my channel in question... 😉 blown!
thanks a lot, for your job and the quality of this content
Thank YOU for leaving a comment, I appreciate the support!
What's youre reason for liking DPM++ SDE Karras above all the others?
This is awesome, man! So helpful! Thanks a lot!
This is unbelivable!
"The final product should be a masterpiece that captures the essence and beauty of the woman" You're giving way too much credit to these prompts. It most definitely just skipped passed 95% of your paragraph you put in there.
I'm also creating a medieval fantasy setting for my game.
thanks for all your tips
I hope they improve hand generation
Can I use realistic vision if I am using Google Colab?
Awesome photos! Thank you for sharing. The hair always looks kind of fake I wish that could be fixed.
Hi Binks. I just learned about Stable Diffusion stand alone app today and it's overwhelming. Is there a place you know where I can learn how to use this app? Also, I've watched several videos and went to several AI sites to see artwork and for those featuring realistic people, the persons are almost always Caucasians. Are there any AI apps that produce people who are Hispanic, Asian, Indian or Black or do I need to prompt the AI to create people of other races than Caucasians? Thanks.
Thank you very much, I have been greatly inspired by your detail explanation. Do you know if the program can run on Mac?
Can you make it use a specific face for all generated pictures?
Thanks for the tips, I was wondering if you can tell me about training the ai more. Because I still don't get it fully.
Thanks alot for this but as soon as I change the prompt to try and get a full body picture, it goes completely wrong, do you have any tips on how to get full body pictures and also where does the model need to be installed.
Thanks
Why when make some character in 1080 there are generate 3 character in 1 image?
Can you try anime characters using ai realistic to like real human?
how to get to the website page that you are showing ?
I get redirect on civitai front page
Can you give info about how to open that page from the civitai front page
對不起,但是您能否親自向我詳細解釋一下,為什麼您無法盡快上傳 #孕婦 #AI繪圖,好嗎? 就如我個人而言,一如既往,我一直相信只要它們在所有年齡段的可接受範圍內,它們就是應該沒問題; 另外,最重要的是,您可以盡快查看 UA-cam 和 Twitter 上所有目前存在的搜索結果,以供您未來的個人參考資料,拜託。
Excuse me, but so could you please explain to me in detail personally, why you can't upload the #AIArts of #PregnantWomen asap, please? Personally, as always, I've believed they should be fine, as long as they've been within the acceptable range for all age groups; also, on top of that, you could go check out all the currently existent search results on UA-cam and Twitter asap, for your future personal reference data, please.
Where can I download Stable Diffusion? Or is it a website? I can't find it
I wrote an extensive reply but youtube ate it. They don't believe in helping or education apparently. I wrote it like 3 or 4 times. It keeps getting ate by youtube.
Just goes to show you that censorship, especially done by AI which creates false positives, is oppression. It is oppression and wrong. UA-cam should be ashamed. I tried to help and I'm sorry that I was unable to.
Friend, to use these prompts I need to have something specific installed in my stable diffuuion or specific configuration?
Why when after I reinstall automatic 1111. The realistic vision 2.0 model checkpoint doesn't give me crisp and clear images in txt2img as before.
it's gives me a blurry & noisy picture even if i put in negative prompt some words to avoid blurriness but it's keep give me less image quality. please any solution?
Have you tried making P.OC. with this model? I have found that A LOT of models either struggle with or just cannot make people of African or Asain desent.
Using the right negative prompt helps alot
negative prompts such as caucasian help a looooooot
Hey man, how long does it usually take to generate? because I press generate and nothing happens, the generate button just changes to 2 buttons, Interrupt and skip.. I have a 3060RTX GPU and heard it should only take 6+ seconds to generate.. so I am unsure why it's not doing anything, can anyone help?
Where is the GitHub repository link?
There's no way to install this for Mac?
Does this work on android
can somebody give me prompts for drawn images cartoon tipe? I have been trying but the images keep looking really crazy ugly.
Can i get this results with my model (.cpk)?
When i use 'Restore Faces' generated image turns blue & green in color please help me🙏
Just don’t use it lol
how did you get into dark mode?
I recently tried quite a few mobile apps, but the usability and result are often dismal.
Thus I"d like to ask what tool, website, pc program or app are you using?
And can you recommend any thing else?
He’s using automatic 1111
i am a newbie at this, what is the reason for batch count 2?
you get 2 pictures instead of 1. you can get up to 50 or 100 different results at a time, but it takes much longer to finish
Very respectfully I am Asking !!!!
I don't know How to get That software and How to Install it. Plz Help Me
I am very Eager to learn and Generate AI art.
Humble request 🙏🏻
Search up a tutorial
Also how do you keep track of the required input data and the related image. For example if you decide you want to recreate or make variations of an older image?
I found one app that actually does this, which us really really helpful and nice.
But the save all of the images and not just your choice results. Thus with just less than a week the history is already cluttered with about 2000 not yet final, and often junk files. Which defies the good idea since you can't easily browse through your wanted results any more.
I download the model, but how I install the model, please help me, thanks
Add the downloaded model to your SD model folder in your files
What version of Stable Diffusion are you using and where can i get it?
This is actually something called realistic vision
What PC setup do you have?
What GPU did you use?
how did you run this, i have downloaded the file but have no idea what to do with it
Put it in your modules folder lol
Are these results possible on an AMD system when using the CPU to render? Just wondering if quality or results are impacted as well as render time or to my hope that render time is the only difference.
you can change the config for it to work on amd
@@brandnewkutta Yes I can run it, just wondering if gpu vs cpu gets different outcomes.
How Stable Diffusion works
can someone tell me which app is he using?
Stable Diffusion WebUI by automatic1111.
github.com/AUTOMATIC1111/stable-diffusion-webui
❤️❤️❤️❤️
Lol mine comes out with blue skin
ETA: 4 hours. BRuh buy me a new PC
Is this utility commercially available somewhere?
On the webpage it says "For version 2.0 it is recommended to use with VAE (to improve generation quality and get rid of blue artifacts)"
How the hell do I install the required files ? xD
👋
Can this work offline?
Very unlikely because of AI always require access to large and often insane amounts of analyzed data.
So no private person could ever copy those. Also depending on if their data is encrypted or not you would have your hands on the balls of their business model., 😅
So those kind of tools will never work offline.
Think of it like having entire Google maps working offline with your tiny smart phone memory.
Yes. SD can run locally
@@noeru9s You are wrong noeru. Its totally offline.
I don't see that "restore faces" option, can someone help me please?
Test