Prompts For Ultra Realistic AI Images: Stable Diffusion
HTML-код
- Опубликовано: 6 мар 2023
- ✨ Support my work on Patreon: / allyourtech
⚔️ Join the Discord server: / discord
🧠 AllYourTech 3D Printing: / @allyourtech3dp
👾 Follow Me on X: / blovereviews
💻My Stable Diffusion PC: kit.co/AllYourTech/stable-dif...
Today i'm going to show you several examples of prompts for stable diffusion that help you get photo realistic results using InvokeAI.
Stable diffusion model used in the video: civitai.com/models/4559/stabl... 
Наука
Oh my GOD!! These look wayyy better than the SD1.5 & SD2.1 models I'm playing with in Invoke AI. Woo wee - I have a lot to learn, but thanks to you I know where to go to get better models. Thank you SOO MUCH for your easy-to-follow-along approach! Haven't been stumped yet in any of your videos. Much appreciated for that!!
tho, people images are still quite rough around the edges, generated objects, like cars and houses can save a ton of time and artist efforts for small game developers by at least giving an idea for asset design. dope tool.
THX man, much appreciated. Nice channel, just subscribed to it :)
Thanks for the sub!
that is seriously helpful, thank you.
You're welcome!
Wow, really cool! :)
Thank you! Cheers!
Thanks so much!
Thank you!!! I’m so glad you found the videos helpful
Failed to import controlnet! The exact error is [WinError 3] The system cannot find the path specified: 'D:\\1111\\stable-diffusion-webui\\extensions\\deforum-for-automatic1111-webui\\models'. Deforum support for ControlNet will not be activated help me with this plzzzzz
I installed it and everything but where do you get the prompts that you used to generate those high quality images that you have on the right?
I've been sharing those in the discord channel (link in the description)
Good explanations. What's your setup? Very fast generating those images. Mine here, Mac Pro 2019, 48GB ram, Radeon Pro W5700K 16GB , took ~60sec for a simple prompt.. ~sigh
Both of my machines are running an Nvidia RTX 3090 which is the main reason my images generate so quickly.
Already run ,everything work fine,Im newbie and don't know which version is the best.I see there are 4 versions there ....
what about ultra realistic of our own face? I would love to do that but its difficult to find information of that
I made a video for that as well: ruclips.net/video/q7wfFMIcvsU/видео.html
how would you go about adding that checkpoint to the images of yourself? Can you load two checkpoints?
You can train a LoRA with images of yourself, and then add that to any model
@@nachobel haven't looked into lora yet, plan on it tonight
I am noticing with some of the results that there is clipping that, normally, I wouldn't think people would want - such as, clipping of the cat's ears, clipping of the cyber car, the head of the girl in the pool at the beginning of the video .... how do you prompt for your images to not be cut off / clipped ?
I typically add information about the type of shot and even the camera. You can say for example “portrait shot on a cannon d7 with a 35mm lens, f2” (I made that up) but guiding the AI with how you want the shot laid out and even the lighting type works quite well.
@@allyourtechai Thank you - I'll give that a try. Much appreciated.
Why do I get, deformed body, deformed face, even if I use detailed prompts and loras?
I didn't know invoke used different prompt decorators than other SD clients. That fucking sucks. Good to know we're fragmenting the community before we really get it going.
Is there a way to install this on Mac?
You should be able to run InvokeAI on an m1 or an m2 Mac, although I hear it is quite slow
Invoke AI is better than Easy DIffusion?
I'm happy with invoke personally, but really it just comes down to using the software that has the features you need. I wish invoke had control net support for example.
Economically it's cool. Naturally...not necessary. Never was the fantasy called economy a necessity us since we arrived. This will totally bring some extra fun though. Namaste
When invoke support lora
sometimes it work sometimes not
this is not stable diffusion this is involve wtf is that title
Not even close to Midjourney v5 realism
MJ5 didn’t exist when this video was made
I did amazing realistic images with just simple prompts with Microsoft Bing without the need to type lengthy prompts and any additional work, other AIs have struggled greatly to produce realistic images with simple prompts
I know this was a year ago so I'll try to be gentle, but these are not "photorealistic".
At the time this was state of the art. As you know, massive improvements have happened