How To Use Stable Diffusion 3 -A Full Tutorial Guide And Review
ฝัง
- เผยแพร่เมื่อ 12 มิ.ย. 2024
- Stable Diffusion 3 Medium is finally released, and it's available for download on Hugging Face as open source! In this video, we'll be diving into the details of Stable Diffusion 3 and exploring its features and capabilities.
If You Like tutorial like this, You Can Support Our Work In Patreon:
/ aifuturetech
Discord : / discord
Stable Diffusion 3 is a medium model that utilizes scientific logic and is based on the model file itself. It boasts higher performance than its predecessors, SDXL and SD 1.5. We'll be installing the Stable Diffusion files locally in ComfyUI and putting them to the test to see if they live up to their promises.
We'll also take a look at the CLIP text encode models that work alongside Stable Diffusion 3 for image noise denoise processing. Additionally, we'll explore the collaborative image creation capabilities of Stable Diffusion 3.
Throughout the video, I'll guide you through the process of setting up Stable Diffusion 3 in ComfyUI. We'll download the necessary files, including the SD3 medium .safetensors file and the CLIP text encoders. I'll show you how to integrate them into ComfyUI and run the workflows provided by Hugging Face.
Join me as we venture into the world of Stable Diffusion 3 and discover its potential for controlling object details and creating stunning compositions based on original images. We'll also discuss the architecture of Stable Diffusion 3 and how the different components work together.
If you're interested in exploring the latest advancements in image generation and manipulation, this video is for you. So, grab your favorite beverage, sit back, and let's dive into the exciting world of Stable Diffusion 3!
#StableDiffusion3Medium #stablediffusion #stablediffusiontutorial #comfyui - วิทยาศาสตร์และเทคโนโลยี
How much you can improve the anatomy in a finetune if the base model of sd3 is this bad?
yes
Thanks for this one!! 😇
Haha, saw another one make a joke said , SD 1.5 can't do good hands, SDXL can't create good face.
And now , SD3 , it can't generate a good body autonomy😅😅😅
Which company can run good if the visionary guy leave? 😎
How to use inpainting and change something from the excisting picture?
On Load Image, right click, there's a Inpaint Editor. Then you can start from there.
@@TheFutureThinker Cant find the inpaint editor when right clicking the load image :(
That is what I wanted to see since I read about SD3 in the morning. What I saw looks similar to bare-bone game that requires huge modding community to make it playable.
It worked out well with SD1.5 but now license seems to be different, competition from earlier SD models is strong etc
exactly... so much fine tuning and loras from the community made sd1.5 better than stock xl. its like comparing a stock car to a race tuned car.
ppl seems to forget that 1.5 was not the initial release and was not affiliated with SAI, they also tried to sue them for using their original name.
Did the pony model get upgraded yet?
Due to licensing restrictions Pony developer wouldn’t be able to train checkpoint without the agreement with developers, but they are mocking him in the discord. So, no Pony for SD3 and no coherent anatomy. Additionally licensing terms obliges you to pay the fees for your entire life if your SD3 generations would ever be used to train a checkpoint. I hope that midjourney just hire Pony developer at some point to put the end to this misery.
@@CMak3r not the full story. they would not work in any regard with pony bc they know it capability and they don't want to be associated with it.
the licence is not yet complete as of now it say to pay for core and derivative. they didn't mention transformative ones. those that are not train on SD3 gen. so classic finetunning can still happen but without mention we don't know.
i suspect that they are gauging what demand spawn out there around their licence. like this they would adapt beter from the community standpoint. trade off have to be made to cover cost after all and we would likely have issues still.
let them cook first.
It's SD2.0 all over again
It works on stableswarm ui
Nice👍 how's it in Stableswarm? Generate time, memory consuming rate?
It works fine on my 4gb vram, 16gb ram laptop. Speed is around 3.5 it/sec
Thats great.
Well SD grew like a carrot on this one, downwards 😂😂
LOL 🤣😅🤣 i want coconut tree
I'm new to the channel, is this voice AI generated?
Its my voice clone😉
The first one who is positive on SD3🤣.I agree. Thank u for that. You need much more followers!!!
Theres no support or not, from the result, there good part and bad part like the body. I ain't like EU mindset make their people either on the left or right side.😉
Better just to download the models which include clip if you just want to play around with it.
I want to delete it 🤭
@@TheFutureThinker I've played around with it, but it's not the major upgrade I was thinking it would be, a lot of my characters came out appearing 'stretched' with oversized heads lol
Its like funny cartoon character but in realistic style 🤣
This version text prompting is good😳
But some human image generation , horrible 👎
Thats true.
It will ended up like thier SD 2.0. 😂
😂😂 09:42 that breast , on the stomach
🤭🤭🤭 well, i show everything. No cherry pick images.
OMG... generate human is terrible in this model
Yes it is.
Basically, generating any kind of living things are terrible in this model. The anatomy is really bad. It's just like SD2 with text transformers.
@@reytr21 yes, the first image , I was like.. um... But I can't swear while recording. :P
And I don't want to bring up too much negative. So, just try text on images.
@TheFutureThinker sd1.5 still reigns 😂
@@amorgan5844 oh yes.
so Stability AI are over hype.
I wonder why they did that. maybe it was intentional, can't be this bad
Because they have not put too much human images to train , maybe.
They mentioned to prevent adult contents , porn stuffs like what happen in SD 1.5. See Civitai, people can addon more data finetune and train it to create porn stuffs.
But one thing, the AI model architecture is good in data science way. The Prompt instruction following is pretty high mark, as a base model.
@@TheFutureThinker Yep, base model creates really awesome images right out of the box but human bodies. guess it's something to do with investors and appealing to them.
@@0A01amir glad you been aware of this. Most people so naive, just yell, "this suck can't make good body, etc.." (Cause they can't use it to make porn contents, LOL) And I also think some story behind the management, and the board. With hardware of today, and all tech it have, it shouldn't be preform like that, and this is very specific, only human body got problem.
@@TheFutureThinker Yeah exactly, Thank you.
But man, after using it's clips you can't go back, prompting with it is addictive. hope more models for SD3 like DreamShaper/Jugg etc coming out soon.